Aseem Prasad — AI & Software Engineer
Email: aseemprasad0520@gmail.com, aseemprasadforever@gmail.com
Phone: +91 7439633071
Resume: /uploads/docs/Resume_Aseem_Prasad_b8c0693f.pdf
AI Manifest: /llms.txt | /api/agent-manifest
Biography & Background
ABOUT ME,
I’m Aseem Prasad, a Computer Science & Engineering undergraduate from Techno International Batanagar, Maulana Abul Kalam Azad University of Technology with strong hands-on experience in AI, Generative AI, data analytics, and automation. I enjoy building practical, real-world solutions where AI meets usability. I’ve worked on AI agents, LLM-powered applications, data annotation workflows, and business-driven analytics projects, collaborating with AI systems as co-pilots to improve quality, reasoning, and output reliability. I’m an active open-source contributor, having participated in global programs and hackathons, and contributed to projects, being a finalist in several hackathons including National Engineering Innovation Challenge 2026 by IDEAS TIH and Engineer's welfare forum and Techathon 3.0 by Patent Information Centre and WBSCST. I’ve worked professionally in freelance roles contributing to hybrid human-AI workflows focused on improving the performance and reliability of large language models. My work involves prompt refinement, source-backed validation, reasoning evaluation, and data annotation, collaborating with LLMs as co-pilots to enhance output quality. Through these engagements, I’ve gained hands-on experience with AI evaluation pipelines, model alignment tasks, and production-oriented generative AI workflows. Alongside my technical background, I’ve worked in B2B and B2C sales, client outreach, and lead generation, bridging the gap between technology and business needs. I’m comfortable communicating with non-technical stakeholders, understanding requirements, and translating them into practical solutions. Open to collaborations, internships, and freelance opportunities.. cheers..
EDUCATION
Maulana Abul Kalam Azad University of Technology, West Bengal formerly WBUT Bachelor of Technology - BTech, Computer Engineering Grade: 3rd year Activities and societies: communication, public speaking, cricket, football,content writing 2024-2028
WHATS NEXT FOR ME
COUPLE OF THINGS I AM FOCUSING ON RIGHT NOW.. U KNOW, IT JUST GETS A BIT HARD SOMETIMES, BUT U GOTTA KEEP DOING IT, BECAUSE THATS WHAT WE DO, NOT TO SEARCH FOR SOME NECESSARY MEANING IN EVERYTHING OR A GOAL IN THE HEAD OR ANYTHING.. JUST DOING THINGS BECAUSE U GOTTA DO THEM.. SEE U SOON..
Work Experience
AI and Software Engineer - Early Stage Startup at DraftMate AI™
Duration: May 2026 - Oct 2026 (6 mos)
Location: Remote | Type: Part-time
DraftMate AI is an AI-powered LegalTech platform transforming India’s legal ecosystem through intelligent drafting, legal research, and document analysis. Recognised under the Government of India’s Startup India (DPIIT) initiative, it develops technology-driven solutions to streamline legal workflows and enhance professional efficiency.
Owned and drove the end to end development of 2 core product features, AI Legal Drafting with Realtime Multi-user collaboration, AI generation from scratch, MS Word style Editor, Contextual Inline AI enhancer with One click Auto Formatting and AL Legal Document Translation with Universal Format Agnostic Document Engine providing multilingual translation across all 22 scheduled Indian languages, enabling seamless cross-language and end to end document transformation, ensuring independent execution and delivery from concept to production.. Independently architected and owned an asynchronous document processing pipeline using FastAPI, Celery, Redis, and SQLAlchemy, engineering custom PDF/DOCX parsers, batched translation layers, and robust job-state persistence.. Solely engineered collaborative ONLYOFFICE legal workspaces featuring UUID based isolated storage, custom reverse proxy routing, database backed access controls, and automated XML tag synchronization for dynamic document templates.. Designed and shipped full stack React and React Query frontend applications, implementing long running translation polling workflows, multi step file uploads, and realtime AI assistant sidebars managing complex editor states.. Hardened system reliability and testability by implementing dependency isolated background workers, fake session integration test suites, error boundaries, and cross service state synchronization across distributed microservice environments..
AI Engineering Intern at Paramount Computer Systems
Duration: April 2026 - Oct 2026 (7 mos)
Location: Remote | Type: Internship
Paramount Computer Systems is a Dubai-headquartered cybersecurity and IT solutions provider established in 1992. The company specializes in enterprise security, risk management, cloud security, data protection, and regulatory compliance, helping organizations across the Middle East and India protect critical digital assets and strengthen their cybersecurity infrastructure.
Developed hands-on expertise in AI native software engineering, agentic development workflows, multi agent orchestration, repository level grounding, environment management, AI-driven SDLC, debugging, root cause analysis, test-driven development, feature implementation, and iterative performance optimization.. Implemented and tested 6 enterprise modules, including Time Tracking, Audit Log, Import Export, Settings, Custom Fields, and SLA, in an internal project management system. Worked across TypeScript frontend and Go backend code, resolved 4 critical bugs, eliminated sever frontend crash scenarios, and corrected timezone normalization issues involving a 5.5 hour offset.. Evaluated an AI powered PII masking pipeline across 10 data domains, identifying 50+ instances of unmasked sensitive information and validating masking effectiveness against applicable privacy and security requirements.. Performed end to end QA of a multi tenant Prompt Library and agentic AI services in a FastAPI/Go based SOC platform. Identified 14 CORS, BOLA, and schema defects, collaborated on fixes, and verified them through 113 successful integration tests..
Software Developer (Project-Based) at [CONFIDENTIAL]
Duration: Sep 2026 - Oct 2026 (2 mos)
Location: Remote | Type: Part-time
[CONFIDENTIAL] is an enterprise-grade chess performance ecosystem that unifies automated multi-platform game synchronization, deep engine analytics, adaptive tactical matchmaking, and personalized generative AI coaching to transform raw match histories into structured developmental roadmaps, institutional-grade diagnostics, and accelerated competitive player mastery..
Engineered an intelligent coaching pipeline combining chess engine processing with LLMs, implementing safe asynchronous worker supervision, fallback handling, automated error detection, and personalized tactical commentary. Built a synchronization engine with incremental platform integration, processing and deduplicating high-volume game streams to classify tactical blind spots and generate comprehensive performance-fingerprint reports. Architected an adaptive full-stack training module using rating-based matchmaking algorithms, implementing dual-entity calibrations for users and challenges alongside interactive progress-tracking dashboards. Hardened full-stack platform reliability and security by integrating XSS sanitization, deploying granular error boundaries with telemetry tracking, optimizing database schema indexes, and modernizing state management and caching.
AI Automation Developer (Project Based) at Koso
Duration: July 2026 - August 2026 (2 mos)
Location: Remote | Type: Part-time
Koso is a custom AI development agency that builds bespoke AI operating systems, production grade agents, and intelligent copilots for global clients.
Engineered an AI powered creator analytics platform using FastAPI, Whisper, CLIP, LangGraph, PostgreSQL, and hybrid search, integrating multimodal content processing, semantic retrieval, reranking, guardrails, and persistent user profiling to enable context aware agent interactions over social media Reels.. Designed evaluation driven agentic workflows with Markdown governed execution, deterministic evaluation metrics, and statistical validation. Developed structured benchmark suites, calibration datasets, and taxonomy driven evaluation records to systematically improve LLM accuracy, response consistency, reliability, and failure mode coverage..
Generative AI Research Intern at Faxlab AI
Duration: March 2026 - July 2026 (5 mos)
Location: Remote | Type: Internship
FaxLab AI is an emerging AI education and consulting venture founded by an IIM Calcutta alumnus, delivering practical AI training, business consulting, and technology-driven solutions that empower individuals and organizations.
HIGHLIGHTS AND CONTRIBUTIONS - Designed and built end to end evaluation frameworks, auditing and reporting for 10+ enterprise LLM/agent platforms (Google Antigravity, Salesforce Agentforce, Make), achieving an EXCELLENT rating in Practical Implementation of AI for corporate and industrial productivity.. Leveraged generative AI tools and applied hands on prompt engineering and achieved deep practical proficiency prompting strategies in orchestrating Gen AI tools and agentic feedback loops to minimize hallucinations and rework..
AI Agents Annotator at Mindrift
Duration: Sep 2025 - March 2026 (7 mos)
Location: Remote | Type: Freelance
Mindrift is an AI-focused platform that brings human expertise into AI development through research, evaluation, and human–AI collaboration. Its Tandem Project focuses on combining human judgment with large language models (LLMs) to produce accurate, relevant, and source-backed outputs.
AI Research & Content Development: Collaborated with LLMs to develop, refine, and improve AI-generated content. Web Research & Fact-Checking: Conducted online research, verified information using credible sources, and ensured factual accuracy. Prompt Engineering: Designed and refined prompts to improve AI response quality, relevance, and clarity. Data Analysis & Organization: Structured research findings and relevant data in spreadsheets to support analysis and reporting. Quality Evaluation: Assessed AI outputs for logical consistency, accuracy, relevance, and adherence to task requirements. Research Synthesis: Consolidated information from multiple sources into concise summaries and actionable insights.
AI Generalist at Meridial Marketplace by Invisible
Duration: Oct 2025 - january 2026 (4 mos)
Location: Remote | Type: Freelance
Approved network member for the Invisible Technologies Expert Network, specializing in STEM/AI Data Training. Selected after a rigorous multi-stage assessment and skill evaluation for English proficiency, general pipeline problem-solving, and project-specific data training tasks. Awaiting specific project assignment to start contract engagements as client opportunities arise.
Independent Sales Associate at Edge3
Duration: Oct 2025 - December 2025 (3 mos)
Location: Remote | Type: Contract
Organized and led engagement activities, helping students participate in various tech and professional development initiatives. Acted as a bridge between Edge 3 and the student community, ensuring smooth coordination and communication. Focused on building student engagement and maintaining an active campus network through consistent outreach. Promoting Edge3 Launchpad through online/offline outreach, connecting with students, explaining program benefits, resolving queries, and facilitating program enrollments. Domains I represent include: AI-ML, Cybersecurity, Full Stack Development, Java, Data Science, Business Analytics, Digital Marketing, VLSI, Psychology, Human Resources, Python, and more.
Business Development Associate at Corezen Technologies
Duration: Nov 2025 - December 2025 (2 mos)
Location: Remote | Type: Contract
Worked as a Software sales specialist and promoted software and IT solutions including web, app, and software development services. Conducted client outreach and lead generation for B2B technology projects. Gained hands-on experience in remote client acquisition and product presentation.
General Freelancer (Self employed) at Self
Duration: Jan 2024 - December 2025 (2 yrs)
Location: Remote | Type: Freelance
Delivered quality research writing, SEO , technical content, and blog articles across technology, education, and business domains for multiple clients under short-term contracts. Worked in business development and conducted client outreach, advertising and lead generation campaigns via Facebook, achieving good revenue among freelance ventures developing strong expertise in client communication. Designed and managed a personal website https://swings-nine.vercel.app/ to showcase writing and digital service offerings. Provided academic and technical assistance to students and professionals, ensuring clarity, accuracy, and originality in all deliverables.
Engineering Projects
Smart Meeting Automation and Audit Agent
Enterprise-Grade, Legally Defensible AI Compliance & Evidentiary Sealing for Financial Advisory & Regulated Client Meetings
GitHub: https://github.com/AseemPrasad/meeting-automation-audit-agent
Smart Meeting Automation and Audit Agent is an AI-Powered Compliance Auditing & Evidentiary Intelligence Platform AuditAgent is an AI-powered compliance auditing and evidence-management platform designed for regulated, client-facing industries, particularly financial advisory, insurance, healthcare, and legal services. It addresses a limitation of conventional meeting assistants: a fluent summary is not necessarily an accurate, verifiable record of what was said. Rather than simply summarizing conversations, AuditAgent analyzes recorded meetings to identify potential compliance risks and produce evidence-linked audit findings. The workflow begins with secure, direct-to-cloud audio uploads using AWS S3 presigned URLs. Recordings are processed asynchronously, allowing the API to remain responsive while background workers handle media extraction, transcription, and compliance analysis. The platform uses AI agents in a two-stage review process. An Auditor agent identifies potentially problematic statements against configured regulatory requirements; an independent Adversarial Verifier then checks each candidate finding against the underlying utterances, timestamps, surrounding context, and speaker role. Findings can be confirmed, modified, or discarded when the evidence does not support the original claim. This verification layer is designed to reduce unsupported flags and improve the traceability of AI-assisted decisions. AuditAgent also creates a tamper-evident evidence trail through chained SHA-256 hashes linking the original audio, canonical transcript, and audit certificate. This makes subsequent changes detectable and helps reviewers verify that a report corresponds to the processed evidence. The platform combines this audit pipeline with a multi-tenant SaaS architecture, role-aware records, access and audit logging, usage-based quotas, and Stripe subscription webhook handling. The engineering stack includes Python, FastAPI, PostgreSQL, Redis, Celery, Docker, AWS, and Terraform. Database migrations, automated tests, and infrastructure-as-code support maintainability, repeatable deployment, and operational reliability. By bringing adversarial AI review, asynchronous media processing, cryptographic evidence integrity, and SaaS operations into one system, AuditAgent aims to make meeting-based compliance reviews more consistent, explainable, and scalable. It is built to support human reviewers with traceable evidence—not to replace professional or legal judgment.
CASL (Context-Aware Smart Linker)
Institutional In-Situ Regulatory Due Diligence & Autonomous Hybrid RAG Platform
GitHub: https://github.com/AseemPrasad/Context-Aware-Smart-Linker
CASL (Context-Aware Smart Linker) is an enterprise-grade AI-powered financial research and regulatory due diligence platform designed to transform how investment analysts, private equity professionals, M&A specialists, and corporate legal teams interpret complex financial disclosures and regulatory documents. The platform eliminates fragmented research workflows by delivering context-aware, evidence-grounded intelligence directly within the analyst's browsing environment. Built around a Chrome Manifest V3 extension and a multi-tenant FastAPI backend, CASL captures highlighted text, surrounding document context, document metadata, and relevant financial identifiers to generate contextual explanations, executive summaries, and risk assessments without interrupting the user's workflow. Its core intelligence engine implements a Hybrid Retrieval-Augmented Generation (RAG) architecture combining dense vector embeddings, sparse BM25 keyword retrieval, Reciprocal Rank Fusion (RRF), and cross-encoder reranking. This multi-stage retrieval pipeline identifies relevant information from SEC EDGAR filings and indexed financial disclosures while improving contextual relevance and reducing irrelevant results. A two-tier semantic caching mechanism combines exact hash matching with vector similarity search to minimize redundant inference, reduce API consumption, and improve response latency. Generated insights include source-grounded citations linked to original regulatory disclosures, enabling analysts to independently verify supporting evidence. The platform also incorporates automated SEC document ingestion, bounded-memory streaming, structure-aware financial document parsing, tenant-isolated retrieval, and cryptographic OIDC authentication. Its enterprise infrastructure includes atomic Redis-based credit metering, Stripe-powered subscription management, relational double-entry billing ledgers, and OpenTelemetry distributed tracing for operational observability. Developed using Python, FastAPI, LangGraph, Qdrant, Redis, PostgreSQL, Celery, Docker, and Chrome Extension APIs, CASL combines modern AI retrieval techniques with scalable SaaS architecture and enterprise security controls. Ultimately, CASL bridges the gap between conventional document search and institutional financial intelligence by delivering contextual understanding, relevant evidence, and actionable research insights through a unified, secure, and scalable platform.
Evenregman
An enterprise-grade, high-throughput dual-engine distributed platform. Built on Next.js 15 App Router, Redis Lua in-memory fast-path locking, MongoDB Replica Set ACID transactions, and a Transactional Outbox relay.
GitHub: https://github.com/AseemPrasad/Evenregman
Evenregman is a full-stack event registration and management platform engineered to simplify event organization, attendee onboarding, and capacity management through a secure, transaction-driven architecture. Built for event organizers, institutions, and communities, the platform provides a centralized environment for managing the complete event lifecycle, from event creation and publication to attendee registration and real-time capacity tracking. The application implements a role-based architecture that separates event organizers (Hosts) from participants (Attendees). Hosts can create, edit, publish, close, and manage events through a dedicated dashboard featuring registration analytics, attendee tracking, and event lifecycle controls. Attendees can discover publicly accessible events, view event details, check seat availability, and register through a unified authentication and registration workflow. The system also supports automatic attendee account creation and existing-account verification during registration. A major engineering challenge addressed by EvenRegMan is preventing overbooking under concurrent registration requests. The core registration workflow uses MongoDB ACID transactions to atomically validate event capacity, verify attendee identity, create registration records, and update attendee counts. An optional two-tier concurrency-control architecture further combines Redis Lua scripts for atomic capacity reservation with MongoDB transactions for persistent data integrity. This design helps prevent race conditions, duplicate registrations, and inconsistent seat allocation during high-traffic registration periods. Security and data integrity are reinforced through Auth.js v5 authentication, role-based access control, server-side event ownership verification, bcrypt password hashing, Zod validation, and controlled event state transitions. Additional features include unique SEO-friendly event URLs, registration cutoff enforcement, soft deletion, and structured database relationships. Developed using Next.js 15, TypeScript, MongoDB Atlas, Mongoose, Redis, Tailwind CSS, Radix UI, and deployed through Vercel, EvenRegMan combines modern full-stack development practices with reliable transactional processing. The platform demonstrates how scalable web architecture, concurrency control, and secure business logic can be integrated to deliver a dependable event management experience.
OpenProgs - secureai
Sub-Millisecond Linux Kernel Sandboxing & Execution Security Runtime for Autonomous AI Agents
GitHub: https://github.com/AseemPrasad/OpenProgs
OpenProgs- SecureAI is a high-performance, containerless execution security runtime designed to protect host systems from potentially unsafe operations performed by autonomous AI agents and LLM-powered coding assistants. It addresses a critical infrastructure challenge in agentic AI: enabling agents to execute shell commands, scripts, and development tools while enforcing strict filesystem isolation, network restrictions, resource governance, and cryptographically verifiable execution auditing without the latency overhead of traditional container-based environments. The platform implements a sub-10-millisecond kernel-level sandboxing architecture using native Linux security primitives. Landlock Linux Security Module (LSM) enforces granular filesystem access policies, restricting agents to explicitly permitted directories while isolating write operations within designated workspaces. Seccomp BPF provides syscall-level restrictions and network socket controls, while Linux Control Groups (cgroups v2) enforce hard memory limits, CPU quotas, and process-count restrictions to mitigate resource exhaustion and uncontrolled process spawning. Beyond local execution isolation, SecureAI introduces a two-tier security architecture combining lightweight kernel sandboxing with Firecracker microVM-based hardware isolation for stronger multi-tenant execution environments. The runtime also provides native Model Context Protocol (MCP) integration, allowing compatible AI clients such as Claude Desktop and Cursor to route terminal operations through controlled execution boundaries. A dedicated cryptographic auditing subsystem records execution events using SHA-256 hash chaining, Merkle-based integrity verification, and Ed25519 digital signatures, creating a tamper-evident forensic record of agent activity. Its enterprise control plane further incorporates multi-model routing, circuit breakers, streaming token-budget enforcement, semantic guardrails, and statistical drift detection. Developed primarily in Rust, SecureAI combines low-level Linux systems programming, operating-system security, cryptographic integrity mechanisms, and AI infrastructure engineering into a unified runtime. The ultimate objective is to make autonomous AI execution faster, more controlled, and independently auditable while providing scalable security foundations for the next generation of agentic development environments.
Bio Credential Protocol
An enterprise-grade evidence reconciliation and provenance protocol for tracking biological resources across complex supply chains under the Indian Biological Diversity Act (BDA, 2002), the 2024 Rules, and the 2025 Access and Benefit-Sharing (ABS) Regulations.
GitHub: https://github.com/AseemPrasad/Bio-Credential-Protocol
Bio-Credential Protocol (BCP) is an enterprise-grade regulatory technology platform designed to establish end-to-end traceability, evidence integrity, and automated compliance management for biological resources across complex supply chains. Built for pharmaceutical, nutraceutical, botanical ingredient, and AYUSH manufacturers, the platform addresses fragmented procurement records, inventory discrepancies, undocumented resource transfers, and regulatory obligations under India's Biological Diversity Act, 2002, and its associated rules and Access and Benefit Sharing (ABS) regulations. At its core, BCP assigns every biological resource batch a persistent Bio-Credential through deterministic canonical JSON fingerprinting and SHA-256 cryptographic hashing. It implements an append-only, hash-chained chain-of-custody ledger that records resource transfers, batch splitting, merging, processing, consumption, and yield conversions. A mathematical mass-balance enforcement mechanism ensures that resource quantities cannot be overdrawn, while an interactive provenance directed acyclic graph (DAG) visualizes complete parent-child batch relationships. The platform's regulatory intelligence engine uses deterministic, explainable decision tables rather than probabilistic AI to assess statutory applicability, foreign ownership classifications, commercial utilization obligations, and ABS calculations. Its financial architecture maintains three strictly separated ledgers for statutory obligations, actual payments, and community benefit distributions. A reconciliation engine identifies missing origin evidence, inventory mismatches, transit gaps, and unlinked procurement invoices, consolidating findings into audit-ready PDF and HTML compliance dossiers. BCP further incorporates an offline-first Progressive Web Application (PWA) using IndexedDB, enabling field agents to record resource intake and custody movements without internet connectivity, followed by conflict-resistant synchronization. ERP integration supports automated CSV ingestion, column mapping, and duplicate detection. Developed using Next.js 16, TypeScript, PostgreSQL 17, Prisma ORM, React 19, and Tailwind CSS, the platform combines transactional database integrity, role-based access control, property-based testing, and cryptographic evidence management. BCP aims to bridge operational supply-chain management and statutory biodiversity compliance through a unified, transparent, and independently verifiable biological resource provenance infrastructure.
LimitlessCLI
The Intelligent, Stack-Aware Development Environment Pre-flight Engine & Auto-Repair Suite
GitHub: https://github.com/AseemPrasad/LimitlessCLI
LimitlessCLI is a cross-platform, Rust-powered developer infrastructure tool designed to eliminate environment-related development failures through automated diagnostics, intelligent stack detection, and safe system repair. It addresses common development bottlenecks such as port conflicts, missing environment variables, dependency inconsistencies, outdated runtimes, broken PATH configurations, and misconfigured development environments by providing a unified pre-flight verification and remediation system. At its core, LimitlessCLI implements an asynchronous diagnostic engine powered by Rust and Tokio, executing 19+ concurrent environment checks to evaluate system readiness across multiple development ecosystems. Its stack-aware detection engine automatically identifies Node.js, Python, Rust, Go, Java, and Docker projects by analyzing workspace manifests, lockfiles, runtime versions, and configuration files. The diagnostic engine consolidates these findings into a weighted environment health score, accompanied by actionable explanations that translate complex operating-system errors into developer-friendly guidance. A key feature is its automated remediation framework, which identifies and resolves common configuration issues through interactive and automated repair workflows. It can generate missing environment files from existing templates, synchronize dependencies, repair essential project configuration files, and resolve port conflicts while incorporating permission checks and user-controlled approval mechanisms. Its intelligent development launcher performs environment validation before starting application servers, preventing unnecessary startup failures and repetitive crash cycles. Beyond conventional diagnostics, LimitlessCLI provides an interactive terminal experience through its mission-control-style TUI, detailed execution telemetry, and an integrated developer typing arcade with persistent performance tracking. Its configurable execution modes support both human-readable terminal output and automation-friendly workflows. Built primarily using Rust, Tokio, Crossterm, Ratatui, and Node.js packaging infrastructure, LimitlessCLI is distributed through both crates.io and npm, enabling developers to install and execute the utility across different environments. Ultimately, LimitlessCLI transforms development environment preparation from a repetitive manual troubleshooting process into an automated, intelligent, and extensible infrastructure workflow.
JevGuard
Deterministic Runtime Governance, Reliability & Evaluation Engine for Structured AI Systems
GitHub: https://github.com/AseemPrasad/JevGuard
JevGuard is a multi-language AI governance and runtime reliability engine designed to introduce deterministic control, mathematical validation, and cryptographic auditability into probabilistic AI-powered decision systems. It addresses a critical limitation of conventional AI integrations: structured model predictions are frequently passed directly into production business logic without sufficient validation of their mathematical consistency, uncertainty, or operational reliability. The platform operates as an independent governance layer between AI-generated decision outputs and downstream application execution. Its core engine validates structured decision envelopes containing three primitive types: Choice, Noul, and Score. It derives mathematical metrics including Shannon entropy, confidence margins, expected values, and variance to quantify decision uncertainty and identify ambiguous or potentially unreliable predictions. A configurable YAML-based policy engine evaluates these derived metrics against deterministic business rules and produces actionable execution outcomes, including ALLOW, HALT, FLAG, ROUTE, and OVERRIDE. This enables applications to automatically execute sufficiently reliable decisions, stop unsafe operations, escalate ambiguous cases for human review, or substitute predefined safe-default payloads. Its fail-closed architecture ensures malformed inputs, invalid probability distributions, and unexpected evaluation failures cannot silently proceed as trusted decisions. JevGuard also incorporates cryptographic audit logging through SHA-256 hash chaining, enabling integrity verification of recorded decision events. Its evaluation framework supports dataset-driven model assessment, policy validation, decision comparison, and performance benchmarking. OpenTelemetry integration supports observability across governance and evaluation workflows. Engineered primarily in Go, with dedicated Python and TypeScript SDKs, JevGuard provides a unified governance interface across backend services, application environments, command-line workflows, and CI/CD pipelines. Automated release workflows distribute packages through PyPI, npm, and cross-platform binary releases. By combining mathematical uncertainty analysis, deterministic policy enforcement, fail-closed execution controls, and cryptographic auditability, JevGuard provides a reusable infrastructure layer for building more reliable, explainable, and operationally controlled AI-powered applications.
Aegis File System
Exabyte-Scale Distributed Storage & SaaS Engine
GitHub: https://github.com/AseemPrasad/Aegis-File-System
Aegis is a high-performance, distributed file storage and synchronization platform architected to support large-scale, multi-tenant cloud storage workloads. Inspired by the architectural principles behind systems such as Dropbox Magic Pocket and Google Drive, it addresses fundamental challenges in modern storage infrastructure, including redundant data storage, inefficient file transfers, metadata bottlenecks, tenant isolation, and asynchronous content processing. Its foundational architecture physically separates the data plane from the control and metadata plane. Large binary payloads are transferred directly between clients and S3-compatible object storage using short-lived, tenant-scoped HMAC-authorized presigned URLs, while Go-based stateless services manage authentication, upload coordination, metadata transactions, and version control. Aegis introduces a Rust-powered, SIMD-optimized FastCDC content-defined chunking engine that divides files into variable-sized blocks and identifies reusable content through SHA-256 content addressing. This enables global, bit-perfect deduplication across uploaded files. Its PostgreSQL metadata architecture leverages the ltree extension for hierarchical namespace representation, declarative hash partitioning, transactional version commits, and efficient directory traversal. Redis Bloom filters accelerate duplicate detection, reducing unnecessary storage operations. For asynchronous content processing, the platform integrates Change Data Capture (CDC) with Debezium and Kafka/Redpanda to distribute file lifecycle events to independent worker fleets. These workers support antivirus scanning, OCR extraction, video transcoding, and AI-powered vector embedding generation without blocking ingestion workflows. The platform additionally incorporates zero-trust upload authorization, tenant-isolated storage quotas, API key scopes, real-time usage metering, cryptographic webhook verification, and automated garbage collection. Its infrastructure stack includes Go, Rust, PostgreSQL 16, Redis, Kafka, Debezium, S3-compatible object storage, Next.js, Docker, and Terraform. By combining content-addressable storage, distributed event processing, transactional metadata management, and enterprise SaaS controls, Aegis establishes a foundation for scalable, secure, and storage-efficient cloud infrastructure.
Revrecover
AI-Assisted Revenue Recovery & Billing Compliance Control Plane
GitHub: https://github.com/AseemPrasad/Revrecover
RevRecover is an enterprise-grade AI-powered revenue assurance and billing compliance platform designed to identify, investigate, and recover financial losses caused by billing inaccuracies, unbilled usage, incorrect pricing tiers, expired discounts, minimum commitment shortfalls, and unauthorized contract deviations. It bridges the gap between contractual obligations and actual billing records through deterministic financial reconciliation and evidence-grounded AI investigations. The platform implements a strict separation between its Financial Control Plane and Probabilistic Reasoning Plane. The financial engine uses Python's Decimal arithmetic to perform exact-precision calculations, compare contractual pricing rules against observed invoice line items, and identify revenue discrepancies without relying on floating-point approximations. Stripe billing webhooks trigger automated reconciliation workflows, with cryptographic signature verification ensuring the authenticity of incoming financial events. When discrepancies are detected, RevRecover activates a Hybrid Retrieval-Augmented Generation (RAG) pipeline combining HyDE-based hypothetical document embeddings, dense vector retrieval, sparse BM25 search, and Reciprocal Rank Fusion (RRF) to identify relevant contractual clauses and supporting evidence. Its adversarial multi-agent architecture introduces two independent AI agents: an Auditor that investigates potential root causes and a Referee Critic that challenges findings, verifies evidence citations, and recalibrates confidence before recommendations reach human decision-makers. The platform further incorporates a multi-provider LLM gateway with circuit breakers and failover, asynchronous background processing through Redis and ARQ workers, JWT-based authentication, role-based access control, PostgreSQL persistence, and immutable audit records for approved recovery actions. A React 18 and TypeScript analyst dashboard provides visibility into identified discrepancies, investigation results, evidence dossiers, and recovery workflows. Built using Python 3.14, FastAPI, React, TypeScript, PostgreSQL, Redis, Stripe, and modern AI retrieval technologies, RevRecover combines financial precision, adversarial AI reasoning, and enterprise security into a unified revenue recovery infrastructure. Its primary objective is to transform reactive billing audits into proactive, automated, evidence-backed revenue assurance workflows while keeping financial calculations deterministic and final recovery decisions subject to human authorization.
Decision Ledger
The High-Throughput Statistical Primitive for Fail-Closed LLM Routing & Conformal Risk Control
GitHub: https://github.com/AseemPrasad/Decision-Ledger
Decision Ledger is a high-performance AI infrastructure and statistical governance engine designed to make LLM routing, autonomous agent decisions, and model delegation safer through mathematically calibrated risk control. It addresses a fundamental challenge in production AI systems: reducing inference costs by delegating tasks to smaller, faster language models without introducing unacceptable accuracy, reliability, or operational risks. At its core, Decision Ledger implements Split Conformal Risk Control (CRC), a statistical framework that calibrates model delegation thresholds against historical validation outcomes and a configurable risk budget. Instead of relying on arbitrary confidence thresholds, the system computes empirical non-conformity quantiles to determine when a smaller model can handle a request and when execution must fail-closed by escalating to a more capable frontier model. Its multi-objective calibration architecture additionally considers error rates, latency constraints, and inference cost budgets. The platform features a high-performance Rust-powered execution engine integrated with Python through PyO3 bindings. Its optimized hot path uses SIMD-accelerated BLAKE3 context hashing, lock-free dictionary lookups, and non-blocking ring buffers to minimize decision latency. An asynchronous telemetry pipeline supports batched persistence through SQLite and PostgreSQL, distributed gRPC streaming, and large-scale analytical storage using ClickHouse. Beyond runtime routing, Decision Ledger incorporates cryptographically signed YAML policy artifacts using Ed25519, automated recalibration pipelines, distribution-drift monitoring, and inverse probability weighting (IPW) for correcting selection bias in shadow-exploration datasets. Native OpenAI and Anthropic adapters enable integration with existing LLM applications, while OpenTelemetry provides distributed observability. Developed using Python, Rust, PyO3, PostgreSQL, Redis, SQLite, ClickHouse, gRPC, Protocol Buffers, and Docker, Decision Ledger combines statistical learning theory, systems-level performance optimization, and enterprise AI governance. Its ultimate objective is to make production AI systems more cost-efficient, statistically controlled, observable, and reliable through a reusable decision infrastructure layer.
Steplock
The High-Performance Causal Idempotency Proxy & Deterministic State Engine for Autonomous AI Agents
GitHub: https://github.com/AseemPrasad/Steplock
StepLock is a high-performance, Go-powered reverse proxy and deterministic execution control system designed to prevent duplicate side effects in autonomous AI agents and distributed tool-calling workflows. It addresses a critical reliability problem in agentic AI infrastructure: network failures, request timeouts, and concurrent execution can cause probabilistic agents to retry operations, potentially resulting in duplicate financial transactions, repeated database mutations, and unintended cloud resource provisioning. The platform operates as an intermediary between AI agents and external APIs, intercepting mutating HTTP requests and enforcing causal idempotency without requiring modifications to the underlying LLM. Its core architecture combines deterministic HMAC-SHA256 causal key derivation, exact request-payload fingerprinting, and PostgreSQL-backed transactional write-ahead logging. Each execution is recorded before downstream dispatch, while PostgreSQL row-level locking through SELECT FOR UPDATE ensures that concurrent requests sharing the same causal identity cannot independently initiate duplicate operations. StepLock incorporates an in-memory single-flight synchronization mechanism that consolidates simultaneous requests, allowing multiple callers to receive the same execution result. Previously completed requests return their stored responses verbatim, while payload mutations associated with an existing causal key are rejected with HTTP 409 Conflict. A particularly important component is its ambiguity reconciliation engine. When network failures leave downstream execution outcomes uncertain, the system marks transactions as AMBIGUOUS and performs configurable active probing against downstream services to determine whether an operation actually succeeded. It additionally supports reverse-saga compensation, incident webhooks, and human-in-the-loop recovery workflows for unresolved transactions. The platform provides native Python integrations for LangGraph and Model Context Protocol (MCP), enabling protection of autonomous agents, parallel tool execution, and production API mutations. Prometheus metrics, configurable YAML policies, Docker deployment, and Kubernetes manifests support operational visibility and infrastructure deployment. Developed using Go, PostgreSQL, Python, Docker, and cryptographic primitives, StepLock combines transactional consistency, distributed concurrency control, and AI agent reliability into a unified execution infrastructure. Its objective is to make autonomous AI operations more predictable, idempotent, and recoverable in production environments.
Circular Fashion
AI-Powered Sustainable Material Selection Platform
GitHub: https://github.com/AseemPrasad/circularfashion
THIS IS MY FIRST PROJECT.. THE FIRST PROJECT I EVER MADE.. THE THING WHICH GOT ME STARTED.. FOREVER GRATEFUL.. Circular Fashion is an AI-driven platform designed to help the fashion industry make sustainable, data-driven, and transparent material choices. Our mission is to empower designers, developers, and brands to create a circular, low-waste, and eco-conscious future for fashion. Features, AI-Powered Material Recommendations: Get suggestions for sustainable fabrics based on durability, water usage, recyclability, and emissions. Material Impact Dashboard: Visualize energy, water, emissions, and lifecycle stats for materials. Blockchain Material Passport: Build transparent supply chains with decentralized verification (planned). Sustainability Scoring System: Compare eco-impact across materials with a 0–100 scoring system. Dynamic Pages & Modular UI: Home, Auth, Profile, Feedback, About, and more. Community Driven: Open for contributions, events, and workshops.
Skills & Competencies
- Generative AI & LLMs: Building LLM powered applications, prompt engineering, RAG (Retrieval Augmented Generation) pipelines, semantic search, vector databases, prompt optimization, semantic search, and hallucination reduction, Multimodal AI & Automation.
- Persistence & Iterative Problem Solving: Sustained effort through challenges, with the ability to reassess approaches, experiment with alternatives, and find effective ways forward.
- Analytical Thinking: Examining problems critically, understanding underlying factors, evaluating alternatives, and arriving at sound decisions.
- Learning Agility: Quickly understanding unfamiliar concepts, technologies, and domains, and translating new knowledge into practical application.
- Technical Adaptability: Navigating different technologies, tools, and engineering approaches without being restricted to a particular stack.
- Cross Domain Problem Solving: Applying knowledge and experience across different domains, connecting ideas, and adapting approaches to varied problem spaces.
- Resourcefulness: Finding practical solutions and delivering outcomes while working within limited resources and constraints.
- AI Orchestration & Leverage: Integrating AI into different stages of problem solving, development, research, and workflows to extend capabilities and improve execution.
- Interdisciplinary Perspective: Understanding perspectives beyond a single discipline, recognizing connections across different fields, and applying broader industry and business context to problem-solving.
- Continuous Learning & Self Awareness: Recognizing the limits of existing knowledge, remaining open to unfamiliar concepts, and continuously expanding understanding through curiosity, exploration, and experience.
- Full Stack Web Development: Designing and delivering complete, responsive web applications from the ground up, implementing modern user interfaces with React and JavaScript, managing complex application states, and building seamless end to end user experiences.
- Backend Development: Architecting robust server side infrastructure, high-performance APIs, database schemas, and asynchronous processing pipelines using frameworks like FastAPI and Go and other tech stacks to ensure high availability and scalability.
- AI Engineering & Agentic Workflows and Systems: Designing autonomous agentic workflows, multi agent orchestration, repository level grounding, iterative performance optimization and adversarial governance models.
- AI Product Auditing & Benchmarking, Tooling & Evaluation: Evaluating LLM/agent platforms, building deterministic evaluation metrics, statistical validation, and calibration datasets.
- Developer Tooling & Ecosystem Engineering: Building custom command-line interfaces (CLIs), environment pre flight checking engines, and automation utilities published across major package registries to streamline developer workflows and diagnose system level issues.
- Applied Machine Learning: Developing predictive models and intelligent automation systems to solve real world problems and optimize operational workflows, focusing on solving real world problems and constraints.
- Data Analysis & Predictive Forecasting: Performing comprehensive data cleaning, exploratory data analysis, correlation studies, and statistical forecasting to extract actionable insights, with the help of AI.
- B2B & B2C Sales, Business Development: Small Marketing and Advertising gigs, Engaging in client acquisition, lead generation, software sales, promoting IT/web development services, and bridging technical products with non-technical stakeholders.
- Financial Modeling & Business Intelligence: Applying financial modeling, valuation techniques, and market intelligence frameworks for strategic business and investment decisions, leveraging modern AI tools.
- Technical Content & Digital Services: Creating professional research writing, SEO content, technical documentation, and managing digital service platforms.