B.Tech CSE '27 @ CMR University, Bengaluru. I build AI agents, GenAI apps, and backend systems — and I ship.
🔴 Live demo: Wildlife Detection — YOLOv8 + ByteTrack, running on Hugging Face — upload a video, get annotated output with per-species counts.
Applied AI — voice agents, RAG systems, LLM eval harnesses:
| Project | What it is | Stack |
|---|---|---|
| VoicePilot | Voice agent: speech → LLM reasoning → speech, 43 tests in the suite | Python, LangChain, LangGraph, FastAPI, Docker, Gradio |
| GroundedQA | RAG + evaluation harness, citation-grounded answers | Python, FastAPI, pgvector, hybrid retrieval |
| EvalForge | LLM evaluation harness, CLI + HTML reports, 21 tests in the suite | Python, deterministic mock-gated CI |
| wildlife-detection | Real-time animal detection, live per-species counts, 28 tests in the suite | YOLOv8, ByteTrack, Pandas, Streamlit |
Backend / Platform — APIs, integrations, identity:
| Project | What it is | Stack |
|---|---|---|
| ProjectFlow | Kanban task-tracking API, 33 tests in the suite | Django REST Framework, PostgreSQL, Docker |
| Green Grid AI | EV-charging site intelligence for Bengaluru, 210K+ records, 17 zones | Python, Pandas, SQL, Google Maps API |
| apertureID | SCIM 2.0 identity provisioning gateway | FastAPI, JWT, RBAC |
Flask, repo-evidenced — built PRD-first, verified, published:
- flask-ledger-api — expense ledger REST API, JWT auth, SQLAlchemy, 34 tests in the suite
- flask-async-jobs — background job service with idempotent submission, 23 tests in the suite
- flask-ml-serving — validated, cached, rate-limited ML inference endpoint, 24 tests in the suite
Newest — debugdost: paste a Python traceback, get the root cause and fix steps in English, Hindi, or Kannada. FastAPI + one-page UI, 5 tests in the suite, live demo on Hugging Face.
- stripe-webhook-idempotency — exactly-once webhook ledger, 36 tests in the suite
- aiprise-vendor-normalizer — 3 KYC vendor APIs → one risk schema, 23 tests in the suite
- plantdoc-grounded-qa — citation-grounded RAG, 92% citation precision, 32 tests in the suite
- agent-harness-eval — baseline vs chain-of-thought agent benchmark, 28 tests in the suite
- indic-voice-agent-eval — Indic speech-to-text WER/latency benchmarks, 36 tests in the suite
Shipped in public, tests in the suite, no private claims nobody can check. Each flagship repo has an Architecture section in its README showing how the pieces fit.
Experience: Backend/API internship at JB Arrow Star (SCIM 2.0, OAuth 2.0, REST).