I build multi-agent and RAG systems end to end: experiment harness, typed production backend, operator dashboard. Five projects below show how I work.

enterprise-ai-customer-support — Adaptive Multi-Agent Customer Support System A LangGraph-orchestrated, memory-augmented multi-agent pipeline (7 typed agents: Intake, Planner, Executor, Critic/Replanner, Memory Manager, Response, Escalation) with a research track (paper-citable ablation results comparing memoryless vs. static-ReAct vs. memory-augmented vs. policy-memory baselines) and a production track (FastAPI backend, per-client auth, OpenTelemetry tracing, Stripe/Zendesk tool integrations, PII redaction) plus a 15-page React operations dashboard with a live LangGraph execution-trace visualizer. LangGraph FastAPI React ChromaDB OpenTelemetry

GroundedRx — Bilingual Medical RAG with a Groundedness Gate A fully self-hosted (no external LLM API) medical RAG system that answers medication questions in Arabic or English from a corpus of patient information leaflets, using hybrid dense+BM25 retrieval and a locally-hosted Qwen2.5-7B model. Every answer is checked against its retrieved source before being shown — the groundedness gate fails closed and has caught a real fabricated answer during evaluation. Cross-lingual retrieval (Arabic query → English source) is measured, not assumed. RAG LLM Qwen2.5 Hybrid Retrieval Arabic NLP

Smart_Reco — Agentic Behavioral Recommendation Platform A LangGraph agent (should_trigger → analyze_interest → retrieve → evaluate_retrieval → refine_query? → generate_recommendation → persist) that watches user activity and recommends courses, gated so most page loads cost zero LLM tokens. Grounding is enforced in code, not just prompted for — persist() filters model output against the actual retrieved product set before writing, so a hallucinated ID never reaches storage — backed by a transactional outbox pattern for Postgres→Chroma sync, a Redis lock to stop duplicate agent runs, and 66 mocked tests plus a separate live-retrieval smoke test that seeds real embeddings and checks top-3 recall. FastAPI LangGraph Chroma PostgreSQL Redis

Kamai-Financial-Companion — AI Financial Companion for India's Gig Workers A 9-agent FastAPI pipeline (budget, tax, risk, savings, and more) where every number — tax slabs, debt-to-income ratios, volatility forecasts — comes from deterministic Python formulas, not the LLM; Gemini/Groq are used only to narrate already-computed results, a deliberate choice since a hallucinated tax figure is a compliance problem, not a UX bug. Supabase Row Level Security enforces per-user data isolation, every backend endpoint verifies a Supabase-issued JWT against JWKS, and a 29-test pytest suite covers the deterministic finance calculations under CI. FastAPI React Supabase PostgreSQL Gemini

predictive_Intelligence — Predictive Intelligence 360 An agentic RAG-powered predictive maintenance platform for industrial equipment (CNC, bearings, pumps, turbofan engines). A 3-stage pipeline — XGBoost fault classification/RUL regression, FAISS semantic search over 1,527 knowledge chunks, and LLM-based diagnostic reasoning — served through a FastAPI backend and real-time dashboard. Backed by a 112-test suite (111 passing) covering ML validation, RAG retrieval, and API correctness. XGBoost FAISS FastAPI sentence-transformers scikit-learn

JEPA-Med-OOD — Self-Supervised Learning for Out-of-Distribution Detection in Medical Imaging Team project comparing JEPA, MAE, and supervised learning for chest X-ray OOD detection — evaluating not just classification accuracy but each method's ability to flag data it wasn't trained on, a key safety requirement for clinical AI deployment. PyTorch Self-Supervised Learning Medical Imaging OOD Detection

StudyGenie Hub — AI-Powered Study Platform (Peekuthon Hackathon) A collaborative study platform with AI-generated flashcards, quizzes, and document summarization, an AI tutoring chatbot, and real-time collaborative study rooms with shared notes and live chat. React TypeScript Node.js MongoDB

Python TypeScript LangChain LangGraph FastAPI React PyTorch XGBoost FAISS PostgreSQL Redis Supabase

• Building production-track infrastructure for multi-agent systems: memory, tracing, tool-failure handling
• B.Tech, Rajiv Gandhi University of Knowledge Technologies, Andhra Pradesh
• Open to AI Engineer internship roles