Building APIs that carry production weight.

A developer-focused RAG system that ingests a real codebase, its docs, and 9 public incident postmortems — answers contextual questions with cited sources, and triages pasted stack traces against past incidents.

Runs fully offline: local LLM via Ollama, local sentence-transformer embeddings, local Chroma vector store, code-aware chunking with tree-sitter.

A local multi-agent travel planner: a LangGraph supervisor coordinates a booking agent, an activity agent, and a budget critic to turn a destination, budget, and preferences into an itinerary that actually fits the constraints — not just the cheapest option.

Combines vector RAG over travel-guide docs with "vectorless" SQL-based retrieval for flights/hotels, backstopped by rule-based guardrails and a scored eval harness. Runs fully offline on Ollama.

An open-source Python library giving a unified API interface to structured USDA nutritional datasets. Migrated legacy scrapers to the official USDA REST API, raising backend reliability to 100%.

Implemented a custom Trie for local matching across 2,900+ brand names, plus TTL caching to cut search latency.

Cleaned and preprocessed 1,000+ customer reviews, then applied sentiment analysis and topic modeling to surface recurring product-feedback themes.

Visualized sentiment distribution and topic frequency to support downstream product decisions.

A network intrusion detection system that monitors device traffic in real time, analyzing patterns to flag unauthorized access, malware infections, and other suspicious behavior.

Built around a client-server architecture with a dedicated traffic analyzer — the tooling that grew out of the Device NIDS work from my Adani/LIAL security internship.