SELECT STAGE

Projects, grouped by domain. Gold border = flagship work. Click any card for the full case study — most include architecture diagrams.

⦿ ACTIVE QUEST — NOW PLAYING Building autonomous L&D agents @ Sword AI ENGINEER · MAY 2026 – PRESENT
Agentic AI
Autonomous agents, MCP servers, LLM tooling
10 STAGES
★★★ STAGE 1-1 Feb 2026 – Feb 2026

Sybil — AI Business Requirements Document Generator

Agentic AI platform that turns scattered project artefacts — emails, transcripts, specs — into a structured 13-section Business Requirements Document. Built as a GDG Hackathon 2026 submission in 2 days.

Next.js 14TypeScriptFastAPIPython
▸ ENTER STAGE
★★★ STAGE 1-2 Jan 2026 – Present

Manul — Agentic Coding Assistant

A terminal-native AI coding agent with a pluggable multi-provider backend and a 16-tool execution layer — works with local and cloud models alike.

AI AgentsTypeScriptLLM ToolingTerminal UI
▸ ENTER STAGE
★★★ STAGE 1-3 Oct 2025 – Dec 2025

Graphing — AI-Drivable Diagram Editor + MCP

A React Flow diagram editor that an AI agent can drive live over MCP — generate, patch, and auto-layout architecture/flow/ERD diagrams, with a CRDT substrate for human↔agent collaboration.

TypeScriptReact FlowModel Context ProtocolYjs CRDT
▸ ENTER STAGE
★★★ STAGE 1-4 Sep 2025 – Mar 2026

Neural Core — Retrieval as a Control Problem

A retrieval-and-reasoning backbone that treats RAG as a control loop — grade, verify, reflect — over an agent-agnostic blackboard where new capabilities plug in with zero rewiring.

PythonFastAPICorrective RAGSelf-RAG
▸ ENTER STAGE
★★★ STAGE 1-5 May 2026 – Present

Meeting Intelligence Platform

A self-hosted meeting notetaker — a bot joins your calls, records and diarizes them, and ships AI summaries, with database-level tenant isolation and a usage-metered credit ledger.

PythonFastAPIPydantic AISpeech-to-Text
▸ ENTER STAGE
★★★ STAGE 1-6 Jun 2026 – Jun 2026

Idolwild — An Ant-Farm for LLM Minds

A browser ant-farm where a deterministic survival sim runs the world for free and an LLM is rationed down to one typed intent per beat — agents invent gods to explain events the player secretly caused.

TypeScriptthree.jsVercel AI SDKZod
▸ ENTER STAGE
★★ STAGE 1-7 Feb 2026 – Present

Perry — Autonomous Personal AI Agent

A 24/7 personal agent running on a Raspberry Pi that doesn't just answer questions — it reaches out, remembers, and acts across calendar, mail, expenses, location, and health.

AI AgentsAutonomous AgentEdge ComputingTool Use
▸ ENTER STAGE
★★ STAGE 1-8 Jan 2024 – Dec 2024

Enterprise Knowledge Repository

Enterprise knowledge repository enabling semantic search across organizational documents and data.

Knowledge ManagementVector DatabaseAzureSemantic Search
▸ ENTER STAGE
★★ STAGE 1-9 Jun 2020 – Jun 2020

IchiBot

Developed and deployed a versatile AIML-powered Chatbot, excelling in dynamic conversations on diverse subjects.

AIMLTelegramChatbotNLP
▸ ENTER STAGE
★★ STAGE 1-10 May 2026 – Present

Application Copilot — HITL Agent

A human-in-the-loop copilot that tailors resumes and drafts screening answers from a learning memory bank — built around hardened browser automation with safety governors and human review.

PythonFastAPIPlaywrightBrowser Automation
▸ ENTER STAGE
Production ML
Medical AI, BFSI systems, models that survive compliance review
8 STAGES
★★★ STAGE 2-1 Jan 2023 – Dec 2024

BFSI ML Systems — Transaction Intelligence, Underwriting & Fraud

Production ML systems for enterprise banking — transaction intelligence, fraud, underwriting, and next-best-action — delivered for enterprise banking clients across UK and Indian markets.

Machine LearningBFSITransaction CategorizationFraud Detection
▸ ENTER STAGE
★★★ STAGE 2-2 Jan 2026 – Feb 2026

Alchemist — An AutoML That Reasons

An AutoML framework where LLM agents profile the data, propose moves from a constrained action registry, and run a self-healing, paper-grounded improvement loop — across classical ML and fine-tuned small language models.

AutoMLLLM OrchestrationLiteLLMPydantic
▸ ENTER STAGE
★★★ STAGE 2-3 Jan 2024 – Dec 2024

XParser - Enterprise Document Intelligence

Enterprise-grade document parser sold at $100K+ per deployment to healthcare clients in USA and Europe

Enterprise AIDocument ParsingOCRLLM
▸ ENTER STAGE
★★★ STAGE 2-4 Jan 2026 – Feb 2026

AutoTrader — Regime-Aware Quant ML Lab

A two-tier quant-ML lab for NSE equities — 36 regime-specialist models under a GPT-based "checker" (a Decision Transformer + RL ensemble), with honest walk-forward backtesting and a documented win for simplicity.

PythonXGBoostLightGBMPyTorch
▸ ENTER STAGE
★★ STAGE 2-5 Mar 2024 – Jun 2024

Lung Cancer Detection with Vision Transformers

A research project applying Vision Transformer architectures to lung cancer detection from medical imaging, exploring attention-based models as an alternative to convolutional approaches.

Deep LearningComputer VisionVision TransformerMedical Imaging
▸ ENTER STAGE
★★ STAGE 2-6 Jan 2024 – Dec 2024

Medical AI Pipeline

Enterprise-grade medical document processing system with high-precision AI summarization and content generation capabilities.

Medical AIDocument ProcessingAWSAzure AI Foundry
▸ ENTER STAGE
★★ STAGE 2-7 Oct 2019 – Nov 2019

Indian Sign Language Interpreter

A Sign Language Interpreter for Mute or Deaf People

Machine LearningComputer VisionTensorFlowOpenCV
▸ ENTER STAGE
★★ STAGE 2-8 Jun 2024 – Sep 2024

Virtual Try-On POC (GenAI)

Stable Diffusion-based virtual try-on POC for major enterprise retail brand's digital commerce platform

Generative AIStable DiffusionComputer VisionRetail
▸ ENTER STAGE
Open Source
Tools shipped to the community
2 STAGES
Game Dev Origin
Where it all started — Unity, C#, national hackathon gold
3 STAGES
Full-Stack
End-to-end apps, serverless, product builds
5 STAGES
???
NEXT STAGE LOADING…
A new build is always in progress. Check back soon.

EXPERIENCE LOG

Where I've worked and what shipped, in detail.

Sword ⦿ CURRENT

Chennai, India

AI Engineer (May 2026 – Present)

AI Trainer Helper — Autonomous Agentic AI for Learning & Development

  • Building an autonomous AI agent that scans and assesses learner profiles to surface skill gaps and personalise recommendations for corporate trainers.
  • Automating end-to-end training-plan creation, sharply reducing the manual effort and turnaround time for individualised plans.
  • Engineering agentic tool-use so the agent acts directly on the organisation’s centralised CMS — updating plans, assigning modules, managing learner records.
LLMsAgentic AITool CallingMCPRAGPython
Xebia

Gurugram, Haryana, India

Xebia is a worldwide, innovative IT Consultancy Company, operating on a global scale

Software Developer, AI/ML (Aug 2022 – May 2026)

  • Medical AI Pipeline - Architected compliance-grade Medical AI pipeline generating ₹2+ crores revenue within 3 months by automating data extraction from 10,000+ documents with 95%+ accuracy. Built XParser (enterprise document parser sold at $100K+ per deployment) featuring three-layer AI enrichment (OCR + CV + LLM orchestration). Implemented 200+ crafted prompts with custom version control. Deployed via AWS AppRunner/Lambda with Azure AI Foundry backend.
  • XParser - Enterprise Document Intelligence - Engineered enterprise-grade document parser sold at $100K+ per deployment to healthcare clients in USA and Europe. Three-layer AI pipeline (OCR + Computer Vision + LLM orchestration) processing multi-format documents with 95%+ accuracy. Powers medical AI pipeline generating ₹2 crores revenue and integrated across 5+ organizational workflows.
  • Knowledge Repository - Designed a company-wide AI repository by processing ~8K SharePoint files into a vector DB for semantic search; optimized related UI.
  • Generative AI - Computer Vision - Developed Stable Diffusion-based Virtual Try-On POC for Fortune 500 retail brand enabling AI-generated apparel visualization. Implemented image conditioning and prompt engineering workflows preserving garment attributes (90%+ accuracy) with near real-time previews (<3 seconds).
  • Banking AI & ML Systems - Built ML-powered PFM systems for leading UK-based SME lender and India-based retail bank: transaction categorization (92% accuracy, 500K+ monthly txns), NBA recommendations (25% loan uptake increase), fraud detection (30% false positive reduction), and underwriting models (40% faster review time). Developed privacy-preserving GenAI chatbots using pseudoanonymization and Small Language Models (SLMs) for on-premise data privacy.
  • Core Banking Infrastructure - Developed Python-based Thought Machine Vault smart contracts for CASA, Mortgages, Loans, and Credit Cards with POS integration. Managed backend systems supporting 10+ client demos. Built Apache Airflow DAG pipelines for document processing workflows with MLflow integration.
  • Open Source Contribution - Built Chomper: open-source MCP server parsing 36+ file formats with TOON (Token-Optimized Object Notation) reducing token usage by 40% for LLM processing. Published as open source under the MIT license.
  • Other AI/ML - AWS AI workflows, Apache Airflow ingestion pipelines, and Neo4j hackathon data insights.
  • Leadership & Mentorship - Mentored engineers in AI/ML workflows and BFSI smart contract development. Conducted technical interviews for Python, Machine Learning, Generative AI, Android, and iOS roles.

Technical Trainee (Oct 2021 - Aug 2022)

  • Internal Chatbots - Built conversational AI prototypes using Rasa for organizational workflows.
  • Internal Smart Contract Work - Contributed to early Thought Machine (Vault) smart contract development and testing.
  • Blockchain-Style Internal PoC - Managed and coordinated an internal proof-of-concept using distributed-ledger principles for background verification use cases.
  • Learning & Foundations - Gained hands-on experience in Python, MongoDB, Postman, and API integrations while supporting multiple internal engineering projects.
Generative AIMachine LearningAgentic AI SystemsThought Machine (Vault)Smart ContractsAWSAzureLangChainNeo4jFastAPIPythonJavaScript
Packt

Online U.S.A

  • Designed an NLU chatbot using RASA for real-time COVID-19 updates, tailored for Indian data
Achievements/Tasks
  • Leadership: Led a team of interns, fostering a collaborative environment and facilitating skill development.
  • Project Management: Conducted daily stand-up meetings to track project progress and address any issues promptly.
  • Delivery: Delivered the project on time, demonstrating effective project management and leadership skills.
  • Recognition: Recognized for delivering the best project in the given timeframe, reflecting commitment and efficiency.
NLU ChatbotRASALeadershipProject Management
CIACON

Online Delhi, India

  • Spearheaded the development of a Capture The Flag (CTF) game for an international event that engaged ethical hackers globally.
  • Leveraged Unity Engine and Photon Multiplayer to create a real-time and seamless gaming experience.
  • Solo Developer.
  • ▶ Play the WebGL build — the CTF emulator is still live in the browser (enter a shared room name for multiplayer).
Capture The Flag (CTF) GameUnity EnginePhoton MultiplayerSolo Developer
Variety Innovation

Gurugram, India

  • Executed tasks related to robotics and collaborative robots (cobots) within the company.
  • Successfully programmed a robot for precision cutting, ensuring accuracy and efficiency in the manufacturing process.
  • Worked extensively with 3D objects, applying Solidworks 3D for design and modeling.
  • Collaborated with cross-functional teams to integrate robotic solutions into various projects.
RoboticsCollaborative Robots (Cobots)Solidworks 3D
▼ OBSERVATION DECK — THE ARCADE FLOOR IS LIVE PICK A CABINET. INSERT COIN. BEAT YOUR BEST.
GAME SCORE 0 BEST 0