Software Developer · AI Systems · Bhubaneswar, India

Ommkar Bisoi

I build intelligent software
— dependable, production-minded systems that read, reason, verify and act, with humans kept firmly in the loop.

Selected work → Get in touch →
0Software systems built end-to-end
0Industries — BFSI · Health · Insurance · Auto
0Tools & frameworks across the stack
∞Curiosity per token
Scroll
01 — Selected Work

Systems that think,
verify & act.

2024 — 2026
PRJ — 01 🏆 1st Place 01
Agentic AI · Hackathon Winner

Enterprise Agentic Operating System

A multi-agent platform that automates enterprise workflows for passenger-vehicle business scenarios — AI agents that analyse organisational data, coordinate with each other, surface operational insights and support executive decisions. Won first place at an AI × cybersecurity hackathon.

  • Multi-agent orchestration with autonomous task delegation
  • Revenue & operational analytics with an executive KPI dashboard
  • Business-process automation and AI-powered decision support
PythonFastAPIAzure OpenAIFAISSStreamlitPlotly
PRJ — 02 02
Agentic KYC · Computer Vision

TrustLens — Agentic KYC Intelligence

An explainable KYC platform where specialised agents handle document OCR, face verification, compliance & adverse-media screening, and risk scoring — then route each case to approve, review or escalate with a human reviewer in the loop.

  • Face verification with InsightFace + ONNX Runtime; live camera capture over WebSockets
  • Exact + fuzzy identity matching across Aadhaar, PAN, DOB & address fields
  • Explainable 0–100 risk scores, audit reports, batch processing — tested on AMD MI300X / ROCm
FastAPIOpenCVInsightFaceONNXOllama · QwenWebSockets
PRJ — 03 03
AI Observability · FinOps · AWS

LLM Usage, Cost & Traceability Platform

An end-to-end observability data model for LLM fleets — every request tracked by user, session, agent, workflow, department and organisation, with token-level cost attribution across orchestrators and nested sub-agents.

  • Prompt / completion / reasoning / cached-token telemetry with model-specific pricing
  • Distributed traces — trace & span IDs across parent-child agent hierarchies
  • AWS pipeline: CloudWatch → Kinesis Firehose → S3 data lake → Athena / Aurora analytics
OpenTelemetryCloudWatchKinesisS3 · AthenaAuroraLambda
PRJ — 04 04
Document Intelligence · LangGraph

Commercial Insurance Underwriting Copilot

An AI copilot that automates commercial-insurance document processing — OCR, classification, structured extraction and eligibility validation — orchestrated as a LangGraph workflow with a rule engine and human review at every uncertain step.

  • PDF parsing & OCR pipeline with document classification
  • Structured JSON extraction with confidence scoring & review escalation
  • Eligibility rule engine wired into an agentic LangGraph state machine
LangGraphLangChainAzure OpenAIFastAPIOCRStreamlit
02 — Also Built

More from the lab.

Side builds · Research · Infra
LAB — 05◈

Medical Discharge Summary Structuring

OCR pipeline converting typed & handwritten discharge summaries into validated, structured JSON with confidence scores, human-review flags and patient-friendly explanations.

Tesseract · Pydantic · Azure OpenAI · FastAPI
LAB — 06◈

News Aggregation RAG Platform

End-to-end retrieval-augmented generation — news ingestion, chunking, embeddings, FAISS semantic search and cited question-answering over live articles.

FAISS · HuggingFace · Embeddings · Streamlit
LAB — 07◈

MCP Client & Server

Custom Model Context Protocol implementation — JSON-RPC over stdio, tool registration and invocation, enabling LLMs to talk to external tools.

MCP · JSON-RPC · Tool Calling · Python
LAB — 08◈

DelayWise AI

Travel-delay intelligence platform — predictive analytics, route dashboards, risk indicators, maps and a RAG-powered in-app assistant.

Predictive Analytics · Plotly · RAG · Streamlit
LAB — 09◈

AutoCat — Email & Document Processing

AI workflow that classifies inbound emails, parses attachments, extracts structured metadata and auto-organises enterprise documents.

Classification · OCR · Azure OpenAI · Structured Outputs
LAB — 10◈

PPE Detection — YOLOv11

Workplace-safety object detection: custom dataset preparation, YOLOv11 training, EfficientNet comparison, and precision / recall / mAP evaluation.

YOLOv11 · PyTorch · OpenCV · EfficientNet
Bonus Level — Vibecoded

Off the clock, on the grid.

Playable · In your browser
A tiny open world

A retro low-poly open world, vibecoded from scratch into a single HTML file — a procedurally generated island with forests, a village and a mountain temple, a full day/night cycle, and a 40-shard treasure hunt. No engine. No assets. Just math, Three.js and good vibes.

  • Procedural terrain, forests, fireflies & wandering critters
  • Day/night cycle with stars, dynamic shadows & low-poly water
  • PS1-style pixelated rendering with CRT scanlines
  • Fully synthesized chiptune soundtrack — zero audio files
Three.jsWebGLProcedural GenerationWeb AudioSingle File
▶ INSERT COIN — PLAY
03 — Capabilities

The toolbox.

Hands-on · Production-minded

A GenAI & LLM Engineering

Azure OpenAIPrompt EngineeringRAGEmbeddingsFAISSSemantic SearchStructured OutputsContext EngineeringOllamaHuggingFace

B Agentic Systems

Multi-Agent OrchestrationLangGraphLangChainMCPTool CallingHuman-in-the-LoopAI Decision PipelinesWorkflow AutomationExplainable Agents

C Backend Engineering

PythonFastAPIREST APIsWebSocketsPydanticSQLAsync WorkflowsAPI IntegrationC++

D Computer Vision & ML

YOLOv11OpenCVTesseract OCRInsightFaceONNX RuntimePyTorchScikit-learnFace VerificationDocument Intelligence

E Cloud & Data

AWS — S3 · Lambda · API Gateway · CloudWatch · Kinesis · Aurora · DynamoDB · Athena · GlueAzure AI FoundryPostgreSQLVector DBsOpenTelemetryData Modelling

F Product & Craft

System DesignAPI DesignCode ReviewsPRDsStreamlit DashboardsPlotlyGit · GitHubLinuxTechnical Storytelling
04 — The Journey

Experience & education.

Chronological · Honest
Experience
2025 — Present

Software Developer — AI Systems

Tata Consultancy Services · Enterprise AI
  • Building enterprise AI products across insurance, healthcare, banking & automotive domains — backends, APIs and workflow engines
  • Secured 1st place in an internal AI × cybersecurity hackathon with a multi-agent enterprise platform
  • Designed LLM observability, cost-attribution and trace architectures on AWS
  • Shipped agentic document-intelligence workflows with human-in-the-loop review
Aug 2024

Intern — Automated Technical Bid Evaluation (RAG)

NALCO (National Aluminium Company)
  • Designed a RAG system to evaluate contractor technical bids against certifications & work history
  • Built conversation pipelines with Flowise AI chatflows powered by LangChain
  • Explored on-premise deployment on NALCO's internal servers using Ollama & LlamaCPP
Education
2021 — 2025

B.Tech, Computer Science

KIIT University, Bhubaneswar · 8.38 CGPA
Focus: machine learning, NLP, retrieval-augmented generation & transformer fine-tuning.

Certifications

deeplearning.ai — Coursera

Machine Learning, Deep Learning & Natural Language Processing Specializations · IBM Full Stack Developer Professional Certificate (ongoing).

Always On

Self-directed research

Agentic architectures, AI FinOps & observability, GPU inference (AMD MI300X · ROCm · ONNX EPs), quantized local models, MCP.

05 — About

Behind the work.

I'm a software developer who believes the most interesting software right now is software that can reason — and that it still has to be engineered like software: observable, testable and honest about uncertainty.

My work sits at the intersection of solid backend engineering and applied AI: REST and WebSocket APIs in FastAPI, multi-agent systems that delegate and verify, document-intelligence pipelines that turn messy paper into validated data, and observability layers that make every token, trace and rupee of an LLM fleet accountable.

I care about the unglamorous parts that make systems trustworthy — typed schemas and clean API contracts, human review where confidence drops, audit trails everywhere, and code the next developer can actually read.

Base
Bhubaneswar, India
Craft
Backend Engineering · APIs · Full-stack Delivery
AI Focus
Agentic AI · RAG · Document Intelligence · LLM Observability
Stack
Python · FastAPI · LangGraph · Azure OpenAI · AWS
Proudest
1st place — Enterprise Agentic AI hackathon
Principle
Autonomy with a human in the loop
Status
Open to interesting AI problems
06 — Contact

Let's build something intelligent together.

Whether it's an agentic system, a document-intelligence pipeline, or just a good conversation about where AI is heading — my inbox is open.

Email me → GitHub ↗ LinkedIn ↗
ommkar003@gmail.com