Backend + AI engineer turning graph reasoning and multi-agent pipelines into production infrastructure for risk, compliance, and healthcare.
I'm a pragmatic software engineer — I care less about what's trendy and more about what actually ships and holds up in production.
AI-driven risk & compliance systems — graph reasoning over knowledge graphs, multi-agent architectures, and low-context LLM pipelines that let organizations make auditable, explainable decisions instead of black-box ones.
Over 3+ years I've shipped realtime systems and agentic workflows across healthcare, compliance, and AI SaaS, on FastAPI, Next.js, Kubernetes, Docker, AWS, and Azure.
- 🔎 Engineered multi-agent fraud & compliance systems using fine-tuned LLMs and knowledge graphs
- ⚡ Scaled a real-time clinical biosignal platform to 100k+ datapoints/min at <50ms latency, used daily by 100+ doctors
- 🎙️ Built a RAG-driven AI recruiting platform with voice analysis, used by 1,000+ candidates weekly
- 🎓 Led engineering on a learner platform serving 70k+ students
- 📖 Springer ICIDA 2024 — Best Paper Award
Instead of scattering experiments across dead repos, I run AI Forge Lab: one modular hub of production-grade guides and systems, each a standalone submodule others can pull independently.
| Project | What it does | Stack |
|---|---|---|
| OpenEvolve | Self-evolving agents that rewrite and improve their own code | Python, LLMs, process isolation |
| Graph LLMs | Knowledge-graph-grounded reasoning for traceable, non-hallucinated answers | Neo4j, Cypher, Groq |
| Vectorless RAG | Hierarchical retrieval that beats flat vector search on evals | PageIndex, LangSmith |
| Fine-Tuning Guide | End-to-end reference for LoRA, QLoRA, DPO, RLHF | PEFT, bitsandbytes, TRL |
| Auto-Research | Autonomous technical/scientific research pipelines | LangChain, Research APIs |
| MCP Chatbot | Modular chatbot built on the Model Context Protocol | LangChain, MCP, Groq |
| GPT-Scratch | Decoder-only GPT built and trained from first principles | PyTorch, tiktoken |
| Podcast Automate | Script-to-speech pipeline, fully automated | Maya1, TTS, LLMs |
| A2A Guide | Patterns for chaining autonomous agent actions | Examples & guides |
70+ public repos. 9 active research tracks under one roof.
✨ Special mention: ML Graph Learning Handbook
A self-built, static handbook covering classical ML and Graph ML from first principles — searchable HTML reader, MathJax equations, Mermaid diagrams, dark mode, zero build step, zero dependencies.
🔗 Live: https://mlnotes.subhajithait.com/
<
C035
/a>
Open to conversations on AI infra, agentic systems, and hard compliance problems — reach out.