Kavin Thakur
Turning high-stakes problems into production-grade AI systems.
About
- Applied AI engineer building production ML systems — CV, NLP, and LLM infra that ships.
- Visiting Researcher at National Ilan University (Taiwan) — sports analytics on NVIDIA RTX 6000 HPC clusters.
- 9× Hackathon Winner — GDG Cloud, AMD × IIT Bombay, PCI Swasth-a-thon (₹1L prize).
- Harvard Aspire Fellow — Digital Transformation & AI Leadership.
- Tech Lead @ IEEE CIET — 20+ devs, 10k+ concurrent users.
Connect
Coding Activity
Stack
Experience
- Sports analytics pipelines on NVIDIA RTX 6000 HPC — targeting O(N) trajectory prediction.
- Focus: Computer Vision, Edge AI optimization, real-time inference.
- PyTorch
- Computer Vision
- Edge AI
- HPC
- NVIDIA RTX 6000
- Real-time Inference
- YOLOv11
- Research
- Official IEEE-CIET website on Next.js/Supabase — 10k+ burst registrations, zero downtime.
- Led 20+ developers across Agile cycles; Azure/AWS Cloud Architecture workshops.
- AI-powered RAG chatbot for natural-language access to IEEE event archives.
- TypeScript
- Next.js
- Supabase
- PostgreSQL
- RAG
- Azure
- AWS
- Agile
- Team Leadership
- Mentored 200+ students across AI/ML and Cloud; organized 4+ national hackathons.
- Code + architecture reviews for CV/NLP projects. Mentor at Formula Hacks 2025.
- AI/ML
- Mentoring
- Cloud
- NLP
- Computer Vision
- Hackathon Organization
- Selective global program — McKinsey Forward: strategy, analytics, and structured case challenges.
- International cohort of emerging leaders; modules on business transformation and problem-solving.
- Strategy
- Problem Solving
- Analytics
- Leadership
- Global Collaboration
- EDA + preprocessing pipelines improving production ML accuracy.
- Classification pipeline reduced preprocessing time; maintained data quality for advanced analytics.
- Python
- EDA
- Data Preprocessing
- Machine Learning
- Pandas
- Scikit-learn
Education
- BE Computer Science (AI & ML) — CGPA 8.72/10
- Core: DSA, OOP, Linear Algebra, Statistics, Cloud Computing, Soft Computing.
- Python
- C++
- DSA
- OOP
- Linear Algebra
- Statistics
- Cloud Computing
- Minor Degree in Artificial Intelligence
- Focus: Neural Networks, Computer Vision (CV), Natural Language Processing (NLP), Robotics.
- Neural Networks
- Computer Vision
- NLP
- Robotics
- Deep Learning
- Leadership Program Certificate — Focus: Digital Transformation & AI
- Social Impact, Innovation, and Leadership.
- Leadership
- Digital Transformation
- AI Strategy
- Innovation
Projects(8)
Graph-RAG pipeline → structured 8-section BRD docs in under 3 min. Top 4 nationally, HackFest 2.0 — GDG Cloud New Delhi.
- Qwen 32B on RTX 4090 — VRAM-constrained inference, asyncio semaphores, 30% speed gain
- LightRAG knowledge graph for entity extraction + automatic stakeholder conflict detection
- Research-backed: Lost in the Middle · Prompt Repetition · Chain of Density
- FastAPI
- LightRAG
- Qwen 32B
- FAISS
- Graph-RAG
- Knowledge Graphs
- asyncio
- VRAM Optimization
- RTX 4090
- Python
Voice AI detecting pathologies with 90%+ accuracy — ₹1L prize, 1st Place at Swasth-a-thon (National Healthcare Innovation, PCI).
- MFCC + spectrogram → interactive clinical reports for early non-invasive screening
- On-device inference (<10s) — zero data leaves the device
- TensorFlow
- PyTorch
- Librosa
- MFCCs
- Spectrograms
- Web3
- Azure
- On-device Inference
- Medical AI
Agentic scheduler cutting coordination time by 70% — AMD × IIT Bombay Finalist, top 1.5% of 4000+ entries.
- Redis caching → 2s latency; high-concurrency vLLM inference
- Python
- Redis
- vLLM
- DeepSeek
- Google Calendar API
- Agentic AI
- High-concurrency
YOLOv11 + IoU + HSV-histogram ReID — persistent player tracking under full occlusion.
- Real-time deployment on NIU HPC clusters (Taiwan)
- YOLOv11
- IoU Tracking
- HSV-histogram ReID
- OpenCV
- PyTorch
- Sports Analytics
- Computer Vision
Production RAG: hybrid FAISS + Redis retrieval, parallel LLM calls, semantic chunking, knowledge-graph integration, async processing, caching, cloud deployment, monitoring.
- FAISS
- Redis
- LLM
- Semantic Chunking
- Knowledge Graphs
- async
- Python
- Docker
IEEE-CIET official site — 300+ live users, 10k+ load-tested, zero downtime. AI-powered RAG chatbot for natural-language event archive access.
- Next.js
- TypeScript
- Supabase
- PostgreSQL
- RAG
- AI Chatbot
- Real-time
- WebSockets
Blockchain-based gamified launch platform using ERC-721/ERC-20 smart contracts on Ethereum.
- Next.js
- Solidity
- Hardhat
- ERC-721
- ERC-20
- Ethers.js
- Web3
- Blockchain
Web2.5 cross-border payments platform bridging traditional finance and blockchain-based settlement.
- Next.js
- Web3
- FinTech
- Cross-border Payments
- REST APIs
Research(4)
Role of Artificial Intelligence and Emerging Technologies in Transforming India's Higher Education
QAHE 2024 — International Conference on Quality Assurance in Higher Education · 2024
Received Best Abstract Award at QAHE 2024. Explored how AI tools are reshaping pedagogy, assessment, and institutional quality systems in Indian higher education.
Revolutionizing Energy Management through Technological Advancements
Research Publication · 2024
Investigated AI-driven optimization strategies for smart grid energy management and demand forecasting.
Real-time Sports Analytics with O(N) Trajectory Prediction
National Ilan University (NIU), Taiwan — HPC Research · 2026
Ongoing research: architecting real-time sports analytics pipelines using PyTorch on NVIDIA RTX 6000 HPC clusters. Targeting O(N) linear-time complexity for trajectory prediction algorithms in elite sports. Focus: Computer Vision, Edge AI optimization, and Real-time Inference.
Independent Research — HackFest 2.0, GDG Cloud New Delhi · 2026
Built and researched a local Graph-RAG pipeline integrating LightRAG knowledge graphs with Chain of Density, Prompt Repetition, and Lost in the Middle papers. Achieved 30% speed improvement on VRAM-constrained Qwen 32B inference on RTX 4090.
Honors & Awards(10)
- Graph-RAG Documentation Intelligence (BRD Agent)
- Competed against 500+ national teams
- VocalWell: AI voice-disorder detection for early, non-invasive clinical screening
- Innovation Prize for Medical AI
- Top 1.5% of 4000+ entries
- AI Scheduling Agent with Redis caching & vLLM inference
- Competitive programming hackathon
- National-level hackathon winner/finalist
- National-level GDG hackathon winner
- Winner at national women-in-tech focused hackathon
- National-level hackathon winner
- 'Role of Artificial Intelligence and Emerging Technologies in Transforming India Higher Education'
- Published and presented at international conference
- Winner at Delhi Technological University hackathon
Certifications(23)
Writing(0)
I was not born with a whole lot of natural talent, but I work hard and I never give up.
— Rock Lee