119,300+ users served by an AI agent I shipped into a $1.4B logistics enterprise
1.08M+ production messages processed, analyzed, and turned into a CTO-level roadmap
50+ universities' worth of course data autonomously enriched by an AWS Bedrock agent
30,546 auto-labeled bounding boxes generated with zero manual annotation
0.995 mAP@50 shelf-intelligence detector shipped for a 200-store retail rollout
7 companies. 1 IIT Madras degree in progress. 0 fear of production systems.
I'm an AI Engineer at Crizac Ltd, currently pursuing a BS in Data Science and Applications from IIT Madras. I build systems that don't stay in a notebook — they ship, they hit real traffic, and they get measured.
My work spans three lanes:
- Agentic AI & LLM systems — multi-agent orchestration, RAG pipelines, and autonomous data pipelines that run unattended at scale (Claude, Bedrock, LangChain, LangGraph)
- Conversational AI at enterprise scale — bilingual, production-grade assistants deployed to hundreds of thousands of real users, extended into internal tools like Microsoft Teams
- Computer Vision & production ML — from zero-labeled-data cold starts to trained, served, and monitored detection systems
I care less about demos and more about the boring, hard parts that make AI trustworthy in production: guardrails against hallucination, human-in-the-loop review, concurrency control, and measuring what's actually happening instead of assuming it.
class AIEngineer:
def __init__(self):
self.name = "Sumit Kumar"
self.role = "AI Engineer"
self.current_company = "Crizac Ltd"
self.education = "IIT Madras — BS, Data Science & Applications ('23–'27)"
self.location = "Goa, India"
self.philosophy = "Ship it, measure it, guard it against hallucination."
def currently(self):
return {
"🏗️ Building": "Agentic AI enrichment pipelines @ Crizac (AWS Bedrock/Claude)",
"🌐 Scaling": "Bilingual conversational AI @ 1.4B-revenue logistics enterprise",
"🔬 Learning": "Advanced agent architectures, graph neural networks",
"🎓 Studying": "Data Science & Applications, IIT Madras",
}🔹 AI Engineer — Crizac Ltd · Goa, India · March 2026 – Present
Architected an agentic AI course-data enrichment platform (Firecrawl → AWS Bedrock/Claude → MongoDB) enriching 50+ universities and tens of thousands of courses across 6 fields with source citations and confidence scoring. Built a concurrent, controllable multi-agent job engine (up to 25 threads/job, semaphore-throttled) cutting enrichment of a 600-course university from hours to minutes. Designed a verified URL-discovery system lifting course-page match coverage from ~2% to 40–89%. Fixed a production bug chain that cut job-startup time by ~700x. Also conducting 30+ technical interviews for AI Engineer roles.
🔹 AI Engineer (Freelance) — The Cloud Intelligence Inc. · Alpharetta, GA (Remote) · February 2026 – Present
Client project: built and shipped "Sultan AI" for SMSA Express — a bilingual (Arabic/English) conversational agent serving 119,300+ users, 127,000+ conversations, and 1.08M+ messages across 325+ cities for a ~$1.4B logistics enterprise. Extended it into a Microsoft Teams custom-engine agent (Bot Framework, Adaptive Cards, SSE streaming, RTL detection) with role-based access via Entra ID. Built the EDA pipeline that quantified failure modes (24.1% dislike rate, 9.3% loop rate) and directly shaped the AI roadmap presented to the CTO.
| Role | Company | Duration | Highlight |
|---|---|---|---|
| ML Engineer Intern | Unsiloed AI (YC F25) | Jan – Mar 2026 | Vision pipelines (Qwen2.5-VL) parsing millions of pages weekly for Fortune 150 banks |
| AI Engineer Intern | ForeignAdmits | Jul 2025 – Jan 2026 | Multi-tenant RAG chatbot + CRM for 5K+ monthly student inquiries; 90%+ accuracy |
| Voice AI Engineer (Freelance) | Hemut (YC W25) | Nov – Dec 2025 | 8 production voice agents automating carrier negotiations |
| AI & Data Science Intern | HTS Tech Solution | Mar – Jul 2025 · PPO received ✅ | YOLOv11 rust detection (85% accuracy) + 3D structural mapping |
| Machine Learning Intern | Unified Mentor | Dec 2024 – Feb 2025 | Fraud detection, medical imaging, and land-cover classification models |
|
End-to-end computer vision system auditing shelf compliance from a single image — detection, SKU classification, counting, and planogram verification. Tech: PyTorch, YOLOv8, OpenCV, FastAPI, MongoDB, React, Docker Impact: 0.995 mAP@50 · 30,546 auto-labeled boxes from zero manual annotation · architected for a 200-store rollout |
Analyzes SEC 10-K filings with a table-aware RAG pipeline that preserves financial statement structure. Tech: FastAPI, Qdrant, Gemini 2.5, React Impact: 2,050+ sections indexed across 89 companies · token usage cut from 150K+ to 15–25K per query |
|
Multi-agent system automating construction quantity takeoffs from architectural PDFs. Tech: LangGraph, RAG, OCR, Computer Vision Impact: Manual takeoff time cut from hours to minutes with human-in-the-loop verification |
Production-grade voice AI platform with Twilio telephony and real-time STT/TTS. Tech: Node.js, Twilio, Groq, Deepgram, Supabase Impact: WebRTC live-call monitoring at sub-500ms latency |
|
Open-source collection of 50 production-ready healthcare AI tools. Tech: FastAPI, Python, Groq, OpenRouter Impact: Full docs, error handling, and deploy-ready templates for real clinical workflows |
Multi-language code quality assessment with RAG-powered codebase Q&A. Tech: FastAPI, FAISS, RAG, MongoDB, AST Parsing Impact: Natural-language Q&A across 10K+ code snippets, GitHub webhook ingestion |
| Domain | Skills |
|---|---|
| 🤖 Agentic & Generative AI | Multi-agent orchestration, RAG, LLM function calling, prompt engineering, LangChain, LangGraph, LlamaIndex |
| 🧠 Computer Vision | YOLOv8/v11, synthetic data generation, domain adaptation, planogram/spatial reasoning, Vision-Language Models |
| 💬 Conversational AI | Bilingual (Arabic/English) NLP, enterprise chat platforms, Microsoft Teams custom-engine agents |
| 🏗️ Backend Engineering | FastAPI, Node.js, REST/GraphQL, JWT/RBAC, multi-tenant systems, async processing |
| ☁️ MLOps & System Design | Concurrency control, rate limiting, model deployment, CI/CD, scalable API design |
| 📊 Data Engineering | Vector databases, ETL, Kafka, large-scale data pipelines |
🎓 BS in Data Science & Applications — IIT Madras (2023–2027) 📰 Published ML research in the IIT Madras Newsletter (Nov 2024) 🥉 3rd place — Gen AI Hackathon by iSource, New Delhi (RAG track) 🥈 2nd runner-up — Syngenta AI Agent Hackathon, Pune 👥 Led 200+ students in a programming community running regular LeetCode sessions 🎤 Facilitated an in-person GenAI workshop for 100+ participants 💼 PPO received from HTS Tech Solution 🎯 30+ technical interviews conducted at Crizac for AI Engineer hiring
🏗️ Building:
- Agentic data-enrichment pipelines on AWS Bedrock/Claude
- Bilingual enterprise conversational AI + Teams integrations
- Computer vision systems for retail shelf compliance
🔬 Exploring:
- Advanced agent architectures (ReAct, Reflexion)
- Graph neural networks & multimodal AI
- Production MLOps and eval-driven LLM development
🎓 Studying:
- BS in Data Science & Applications @ IIT Madras
Open to collaborating on agentic AI, computer vision, and production-scale ML systems.