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

Production-grade AI systems — multi-agent orchestration, RAG pipelines, LLM integrations, and computer vision models that operate under real constraints.

React TypeScript Gemini 2.5 Flash/Pro Supabase Edge Functions

Two-phase parallel LLM architecture: Phase 1 runs Research + Pro/Con Advocate agents via Gemini 2.5 Flash; Phase 2 runs Bias Checker + Fact Verifier via Gemini 2.5 Pro; Synthesizer delivers confidence-scored output with real-time streaming. 3–5s analysis vs 20s+ sequential · 🌐 Live Demo

TypeScript React PostgreSQL TigerDB LLM Agents

4 specialised AI agents (Quality · Security · Performance · Docs) running in parallel with human-in-the-loop review gates via TigerDB zero-copy forks. 4× faster analysis (40s → 10s), zero storage overhead. 1st place — Agentic PostgreSQL Challenge

Python Agno Groq LLM FireCrawl Exa API Streamlit

Autonomous SEO audit agent that crawls sites, runs RAG-powered technical audits, keyword gap analysis, and competitor benchmarking — delivering GDPR-compliant dashboards. +20% client organic performance. 🌐 Live Demo

Python PyTorch OpenCV ResNet-50 EfficientNet-B3 NumPy

End-to-end defect detection system: 500+ images/min · ResNet-50 + EfficientNet-B3 ensemble · MixUp/CutMix augmentation · Test-Time Augmentation · 90%+ accuracy across 12 defect classes · training data expanded 3×.

Python LangChain FAISS Groq LLM Scrapy Streamlit

Agentic pipeline: scrapes job listings → embeds portfolio into FAISS vector store → RAG + dynamic prompt engineering → hyper-personalised cold emails. +25% recruiter response rate.

Zapier WhatsApp API LLM n8n

End-to-end WhatsApp automation agent handling student queries autonomously — no human in the loop for routine support flows. FAQ resolution · escalation routing · session management.

TypeScript n8n Mixtral Webhook

n8n + Mixtral workflow for automated professional LinkedIn post generation — prompt in, polished SEO-optimised content out.

• LLM Automation: Built a Python CLI agent chaining 10+ REST APIs (Azure AD OAuth2) using function-calling patterns, replacing a 15-step manual workflow with one command — 80% faster onboarding
• Computer Vision: Implemented ResNet transfer learning + autoencoder anomaly detection in PyTorch — 90%+ accuracy, feature dimensionality reduced 65%
• Data Pipeline: Processed 10,000+ industrial images through automated augmentation pipelines; expanded training data 3×, improved model accuracy 15% across 5 product categories

• Built 5+ event-driven scraping pipelines (Scrapy · BeautifulSoup · Pandas) processing 100,000+ records
• Docker-containerised CI/CD-scheduled workflows → 60% reduction in manual collection, 10+ hrs/week saved
• Integrated REST APIs with XPath/CSS parsing and fault-tolerant retry middleware

• Top 5% of 7,000+ contributors globally · Pull Shark ×2 · Quickdraw · Starstruck achievements
• Merged 20+ PRs across LLM tooling and agent framework repositories
• Reviewed 30+ peer PRs, mentoring contributors and reducing review cycle time 25%

📦 12-Class Casting Defect Detection Dataset — Kaggle

The only publicly available multi-class extension of the binary casting defect benchmark. Paired with a ResNet-50 + EfficientNet-B3 ensemble using MixUp/CutMix augmentation and TTA — 90%+ accuracy across all 12 defect classes. ORCID: 0009-0002-2854-8468