🎯 Role        →  GenAI / LLM Engineer — Agentic Systems, RAG & Knowledge Graphs
📍 Location    →  Pune, India (Remote-friendly)
🎓 Education   →  B.E. Computer Science, MSU Vadodara — CGPA 8.24 (2023–2026)
🏆 Wins        →  3× Hackathon Winner (Global Agent Hackathon · Agentic Postgres · Bhashathon IIT Bombay)
⚡ Superpower  →  Building agentic LLM systems that ship to production, not just demos
🔬 Research    →  Published 12-class casting defect dataset on Kaggle (ORCID registered)
🌍 Open Source →  Top 5% of 7,000+ contributors globally — GSSoC Extended 2024

Python FastAPI LangChain Redis LangSmith · 2025

FastAPI RAG microservice with an instruction-hierarchy guardrail blocking 95%+ prompt-injection attempts and circuit-breaker fallback; full tracing/eval via LangSmith. Cut p95 latency 45% (2.1s → 1.15s) and tokens/request 35% via caching/streaming, scaling to 50+ req/sec.

Python MCP SDK BM25Plus · 2026

MCP server exposing document indexing and search as agent-callable tools, with BM25Plus-ranked retrieval and parent→child category-prefix scoping over a hierarchical taxonomy; fully offline with no model downloads. Caught and fixed a classic-BM25 zero-IDF edge case on small corpora via a 6-test pytest suite (6/6 passing), and shipped a working Claude Desktop integration config.

TypeScript React PostgreSQL TigerDB LLM Agents · 2025

4 autonomous AI agents (Quality · Security · Performance · Documentation) running in parallel with human-in-the-loop gates via TigerDB zero-copy forks. 4× faster analysis (40s → 10s). 1st place — Agentic Postgres Challenge.

• End-to-End ML Pipeline: Owned the full lifecycle of an industrial defect-detection system — data ingestion, augmentation, PyTorch model training, Azure-hosted deployment, and post-deployment monitoring — taking it from prototype to a production-owned service
• Secure Agentic Integration: Designed a Python function-calling agent with Azure AD OAuth2-authenticated service-to-service integration, orchestrating 10+ REST APIs to replace a 15-step manual workflow and cut onboarding time 80%
• Casting Defect Classification: Trained ResNet transfer-learning and autoencoder models in PyTorch and deployed to production to flag manufacturing defects pre-shipment — 80%+ accuracy, replacing manual visual inspection on the line and reducing missed-defect escapes
• HTS/HSN Tariff Classification System: Built a multi-agent SupervisorAgent + Chapter84Agent classifier on BM25 parent→child code-prefix retrieval and Azure OpenAI, paired with a Neo4j tariff-hierarchy knowledge graph for GRI-based disambiguation; instrumented full tracing and cost/latency monitoring with Langfuse, and benchmarked GPT-5, Claude Opus, and Claude Sonnet across 100+ industrial line items
• Data Pipeline: Engineered SQL-backed image pipelines on Azure infra over 10,000+ samples, expanding training data 3× and lifting accuracy 15%

• Pipelines: Designed 5+ scraping pipelines (Scrapy · BeautifulSoup · Pandas · PostgreSQL) processing 100,000+ records with fault-tolerant retry middleware
• Automation: Deployed Docker, CI/CD-scheduled workflows cutting manual data collection 60%, saving 10+ hrs/week

• Top 5% of 7,000+ contributors globally — merged 20+ PRs across LLM tooling and GenAI repos; reviewed 30+ peer PRs, cutting review cycle time 25%

📦 12-Class Casting Defect Detection Dataset — Kaggle

ORCID: 0009-0002-2854-8468