I build LLM applications and data-science models in Python — RAG pipelines, chatbots and voicebots,
LangGraph agents with guardrails — and the REST APIs behind them, shipped with evaluation and tests.
Bangalore, India
- RAG & retrieval — chunking, embeddings, hybrid search (pgvector HNSW + BM25 with Reciprocal Rank Fusion), LanceDB, ChromaDB, citations and out-of-scope refusal
- Agents & conversational AI — LangGraph agents with relevance gates, tool calling, MCP, voicebots with speech-to-text / text-to-speech, multi-provider LLM fallback
- Evaluation & data science — LLM-as-judge bias measurement, statistical testing, forecasting, scikit-learn models
- Engineering — FastAPI / Express APIs, Next.js frontends, PostgreSQL with RLS, pytest + GitHub Actions CI (500+ automated tests across public repos)
| Project | Highlights | Stack |
|---|---|---|
| lenny-growth-assistant | Grounded RAG over podcast transcripts; hybrid retrieval, timestamp citations, multi-LLM fallback | Python · FastAPI · pgvector · BM25 · Next.js |
| cost-efficient-rag | 0.71 Recall@5, 34.6 ms p95, documented token-cost / TCO comparison | Python · FastAPI · LanceDB · litellm |
| llm-as-judge-pipeline | Measures 5 judge biases against 105 hand-labeled judgments (Cohen's κ, Wilcoxon); 200 tests | Python · scikit-learn · SciPy |
| hotel-bar-inventory-intelligence | Demand forecasting across 96 series; backtest cut stockout days 73–87%, fill rate 28% → 94% | Python · statsmodels · scikit-learn |
| agentic-ai-rag-chatbot | Document-grounded chatbot with a LangGraph gate that refuses below a confidence threshold | Python · LangGraph · Pinecone · FastAPI |
| ai-interview-prep-kit | Job description → requirement-traced interview kit with a deterministic coverage engine, 5 LLM providers | TypeScript · Node.js |
| zoho-one-employee-portal | Own JWT sessions + RBAC, Zoho apps through one server-held account, audit logs, RLS | Next.js · Express · Supabase |
| ats-resume-optimizer | TF-IDF resume-to-job matching with keyword gaps; 276 pytest tests, CI on 3.11 / 3.12 | Python · scikit-learn · Streamlit |
AI: LangChain · LangGraph · Hugging Face Transformers · Ollama · GPT-4o · Gemini · sentence-transformers · Whisper · MCP
- AppSec-Rules-Pack #35 — portable JSON validation-gate example with focused tests, closes #31 (merged)
Open to AI/ML engineering roles — reach me through my portfolio.