I build AI systems that work in the real world — robotics, industrial troubleshooting, clinical decision support, retrieval systems, and conversational interfaces.
My work blends AI, software engineering, and human-centered design.
I specialize in:
- Retrieval-Augmented Generation (RAG)
- Agentic workflows for real tasks
- Model optimization (local inference, llama.cpp)
- Conversational systems & search interfaces
- Evaluation of human–AI decision-making
I care about making AI useful, reliable, and understandable — not just high-scoring in benchmarks.
A full RAG + agent pipeline running inside a Furhat social robot.
- Adaptive + Corrective + Self-RAG
- Sub-8s on-device inference (llama.cpp)
- FastAPI backend on EC2 (<4s generation)
- First agent-driven robot backend in the research group
- Rated Outstanding in my work-term evaluation
AI system for frontline industrial operators.
- First agent shipped within “Atlas AI”
- 90% accurate keyword extraction (embeddings + cross-encoders)
- Gemini multimodal integration for GCP users
- Recursive Tail Generation for long-term memory
- 90% X-ray diagnostic accuracy (CNNs + ViTs)
- RAPTOR-powered retrieval (95% clinician-validated)
- Personalized rehab guidance using medical-text RAG
- Published at an AI conference
- Designing experiments on calibrated human–AI decision-making
- Building conversational search/recommendation systems
- First-author CHI submission
- Austin, Texas — Industrial AI (Cognite)
- Boston, Massachusetts — Ops dashboards & tooling (PathAI)
- Trondheim, Norway — Robotics + RAG research (NorwAI, NTNU)
- Toronto, Ontario — Clinical AI + startup engineering + Banking Quality Automation Engineering
- Waterloo, Ontario — Research + engineering degree
- Vancouver, BC — Early software work and customer-facing roles
These experiences shaped my ability to work across diverse cultures, industries, and technical stacks.
Outside of my AI/ML work, I’ve also built:
- Full-stack MERN applications (startup/fintech)
- Banking automation testing frameworks (TD)
- Database + vector search systems (MongoDB redesigns)
- Predictive analytics tools and dashboards
- Production CI/CD pipelines (Jenkins, Firebase, Supabase)
A lot of my breadth comes from working across startups, industry, research labs, and enterprise environments.
👉 For my full work experience and detailed career history, check my LinkedIn:
https://linkedin.com/in/jeevan-parmar-62b464194
AI / ML: PyTorch, HuggingFace, RAG, Agents, OpenAI, Gemini, llama.cpp
Backend: FastAPI, Node.js, Docker, AWS/GCP/Azure, microservices
Full-Stack: React, Vue, TypeScript, Express, SQL/NoSQL
Infra / Tools: ChromaDB, Supabase, Jenkins, Firebase
Specialties: Model optimization, conversational systems, applied ML research
- Search Engine (BM25 + embeddings): Query-biased summaries + statistical evaluation
- Audio Transcriber: Whisper + GPT cleaning pipeline with HITL workflows
- Meal Stream: Full-stack meal planning + nutrition analytics
- Energy Price Forecasting: ML models for PJM market prediction
- NBA Player Projection: Live model deployment w/ analytics dashboard
I’m studying Management Engineering (AI Option) at the University of Waterloo — a mix of:
- Machine Learning
- Optimization
- HCI
- Software engineering
- Systems design
- Decision analysis
My academic work includes CHI-focused research and building conversational/agentic systems.
I train Brazilian Jiu-Jitsu, run marathons, play football (both), basketball, squash, tennis, and golf.
I also enjoy collecting colognes and watching absurdly long sports documentaries.
- LinkedIn: https://linkedin.com/in/jeevan-parmar-62b464194
- GitHub: https://github.com/jeevanp03
- Email: j29parma@uwaterloo.ca