I'm an AI Engineer based in Ahmedabad, India who builds production AI systems — not prototypes, not demos — real systems that serve real users every day.
In under 2 years, I've shipped 6 end-to-end AI systems across the full spectrum: from soldering IMU sensors onto finger rings for TinyML devices to orchestrating multi-model LLM architectures (GPT-4o + Gemini + Qwen3-Coder) in production.
🔭 Currently: AI Engineer @ Vivansh Infotech — sole AI engineer on an enterprise EHS platform
🏗️ Shipped: 6 production AI systems serving real users daily
📊 Scale: RAG chatbot handling 50K+ queries/month | 500+ concurrent users
🎓 Research: IIT Delhi (wearable AI) | IIT Gandhinagar (embeddings & retrieval)
🏆 Recognition: Samsung Top 10/70K+ | Smart India Hackathon Top 5/44K+ | CNN News18 Feature| # | System | What It Does | Scale | Impact |
|---|---|---|---|---|
| 1 | Permission-Aware RAG Chatbot | Multi-tenant chatbot enforcing 250+ permissions per query in real time | 500+ concurrent users, ~1K queries/day | 80% MRR, ~8s response time |
| 2 | AI Form Builder | Converts PDFs → digital forms in ~20 sec; also builds forms from natural language | 200+ forms in month 1 | 99%+ time reduction (1+ hour → 20 sec) |
| 3 | AI Video Course Builder | End-to-end automated video course production from a single document | 10,000+ video clips indexed | 99% time reduction (1 week → ~10 min), ~$2/course |
| 4 | Production RAG Platform | Enterprise chatbot platform with 10+ simultaneous instances | 50K+ queries/month, thousands of clients | Improved MRR from 45% → 95% |
| 5 | On-Premise OCR Pipeline | Hybrid OCR for sensitive documents (zero cloud dependency) | 200+ page PDFs | Fully air-gapped, government-grade |
| 6 | Voicer — Sign Language Device | Wearable AI translating Indian Sign Language → spoken audio in real time | 94.1% accuracy, 12 gesture classes | Samsung Top 10, CNN News18 feature |
Languages
AI / ML / LLMs
RAG & Vector DBs
Backend & Infrastructure
Hardware & Edge AI
🥇 Samsung Solve for Tomorrow — Top 10 among 70,000+ participants nationwide (Featured on CNN News18)
🥇 Smart India Hackathon 2023 — Top 5 among 44,000+ teams (Hardware Edition, MIT Karnataka)
🥈 Ingenium Hackathon — 2nd Place among 200+ teams
🧩 370+ competitive programming problems solved in Python
📜 Microsoft Azure AI-900 Certified | deeplearning.ai — Neural Networks & Deep Learning
┌─────────────────────────────────────────────────────────────────────┐
│ HARDWARE │ Sensor rigs, IMU rings, 3D-printed enclosures │
│ EDGE ML │ TinyML on Arduino/ESP32 (55KB flash, 2KB RAM) │
│ DATA ENG │ 12,400 gesture samples, 10K+ video clips │
│ ML TRAINING │ 1,000+ model configs, CNN-BiLSTM hybrids │
│ RETRIEVAL │ Production RAG, Milvus, multi-index, reranking │
│ LLM APPS │ GPT-4o + Gemini + Qwen3-Coder orchestration │
│ INFRA │ AWS Lambda, vLLM, on-premise LLM hosting │
│ DOC PROCESSING │ Hybrid OCR (DocTR + DocLink), PDF pipelines │
└─────────────────────────────────────────────────────────────────────┘
B.E. Computer Engineering — Vishwakarma Government Engineering College (GTU), 2020–2024
Research Intern — IIT Delhi (Jan–May 2024)
- Built a wearable air-writing recognition system (62 classes, 12,400 samples, 1,000+ model configs)
- Supervised by Dr. Lalan Kumar (IIT Delhi) & Prof. Amit Rathod (VGEC)
- Collaborated with researchers from IISc Bangalore & Howard University
Research Exposure — IIT Gandhinagar
- Worked with Prof. Mayank Singh on embedding models & retrieval systems
- Foundation for all subsequent production RAG work
| Decision | What I Tested | Result |
|---|---|---|
| Vector DB | FAISS (45% MRR) vs Qdrant (85%) vs Milvus (95%) | Selected Milvus |
| Embeddings | BGE small/base/large, Stella, SPLADE, BGE-M3 | Selected BGE-M3 + reranker |
| Form Generation | GPT-4o (structural failures) vs Qwen3-Coder-480B (perfect accuracy) | Selected Qwen3-Coder |
| Multilingual | Native multilingual models vs translate-bridge pipeline | Translate-bridge won |
| Video Reranking | LLM reranking vs Cohere rerank | Cohere — better coherence |
💬 I'm always open to collaborating on AI projects. Let's build something that ships.