About Rochon
Rochon is an independent AI research and engineering laboratory. We investigate, evaluate, and build intelligent systems for organizations where control, data privacy, and physical hardware constraints matter.
The lab exists because most AI advice is vendor advice. We hold no reseller agreements, take no referral fees, and ship no software we have not physically measured. Clients keep the weights, the code, and the direct invoices from their own compute providers.
Two connected missions
Custom systems in production
We build and deploy private AI systems: air-gapped inference inside your racks, domain-specific small language models, deterministic agent workflows, and low-power embedded computer vision. Client work creates practical value and exposes real edge cases.
First-principles investigation
We investigate how neural systems behave under physical constraints: where they hallucinate, how tokenizers can be adapted for industrial vocabularies, and how quantization affects execution. Over time, we publish open papers, benchmarks, datasets, and tools.
Client engineering exposes hard problems. Independent research builds the technical depth to solve them. Each informs the other.
Five core beliefs
Methodology & approach →The right AI fits
→A small model that knows your vocabulary beats a giant that guesses.
Data belongs to you
→Weights, code and logs transfer to the client. No lock-in, no resale.
Evidence beats opinion
→Every recommendation is backed by a measurement you can reproduce.
Simple wins
→The architecture your team can operate is the architecture that survives.
Honesty about limits
→If AI is the wrong tool for your problem, we say so in week one.
Mehdi Namdar
Founder & AI Systems Engineer
Electronics & Instrumentation Engineering, Nirma University
Mehdi is an AI systems engineer with a background in Electronics & Instrumentation Engineering at Nirma University. His work focuses on the full deployment arc: from tokenizer design, dataset curation, and quantized parameter tuning to hardware-level execution on bare-metal accelerators and low-power embedded silicon.
- TARANG: Edge AI cardiac-monitoring system recognized as a Global Top 10 finalist in the Silicon Labs IoT Challenge.
- ComplianceAI: Air-gapped local LLM retrieval and document-analysis pipeline operating with zero outbound API telemetry.
- StoryHack: Multi-agent orchestration pipeline coordinating 13 specialized agents across manuscript extraction, narration, and execution.
- Robotics & Perception: Computer vision and sensor integration engineering with Team Dyaus.
Deploy custom AI on your terms.
Tell us about your target silicon, data residency constraints, or latency requirements. Rochon engineers will provide a clear evaluation of what is viable, with power and latency models and deployment blueprints.