Projects

An enterprise Document Intelligence system built with LangGraph, Azure OpenAI, FastAPI, and Azure Container Apps. The application processes questionnaire spreadsheets against validation materials, extracting text from PDF, DOCX, and Excel files with optional OCR for scanned PDFs.

The system demonstrates agentic RAG architecture with document validation workflows, routing questions through simple, rating, or deep validation agents. Results are streamed to the UI and written to a downloadable Excel file. Includes production CI/CD with proprietary implementation safely stubbed.

A production-grade Agentic Retrieval-Augmented Generation (RAG) system built with a dual-environment approach: Ollama for local experimentation and GCP with Vertex AI (Gemini) for scalable production deployment. The system follows a structured, state-driven reasoning pipeline — Planner → Retriever → Tool Executor → Synthesizer → Validator — enabling dynamic decision-making, conditional branching, and iterative refinement far beyond static RAG.

In production, an intelligent hybrid LLM routing strategy dispatches simple queries to cost-effective local inference and complex reasoning to Vertex AI (Gemini), achieving 30–70% cost savings. Cloud Run provides auto-scaling, while streaming responses reduce perceived latency by 40–60%. Full observability is delivered via LangSmith API tracing and Cloud Logging/Monitoring.

A GenAI-powered health analytics assistant for chronic kidney disease (CKD) risk, built with Python, LangGraph, LangChain, and Ollama. It integrates patient-level data from multiple health datasets, runs CKD risk modeling using RandomForest and XGBoost, and produces normalised feature-importance scores with SHAP-style explanations.

A LangGraph v2 state graph orchestrates the full pipeline — joining data, planning tool usage, and conditionally invoking analytics, hybrid vector RAG (Chroma + Ollama embeddings), and MCP web search (SearxNG primary, DuckDuckGo fallback) before streaming a final explainable answer through a FastAPI endpoint to a Streamlit chat UI.