This page may contain stale information. Last updated: 2026-06-28
Overview
Mirendil is a San Francisco AI lab (founded early 2026) building infrastructure to automate AI research — experimental design, hyperparameter search, evaluation, and iterative training — for organizations outside major labs.
Founding Team
- Behnam Neyshabur (CEO): Ex-Google Gemini reasoning (5.5 years), ex-anthropic (Dec 2024–Dec 2025)
- Harsh Mehta (CTO): Ex-anthropic research scientist
- Shayan Salehian: Early xai member
- Tara Rezaei: MIT graduate, ex-openai
- ~20 researchers/engineers from Anthropic, OpenAI, DeepMind, xAI
Funding
- 2026-06-26: 1B valuation — co-led by andreessen-horowitz and kleiner-perkins; nvidia participating (2026-06-26-mirendil-unite-ai-seed-round)
Product Thesis
Self-optimizing platform: formulates hypotheses, generates code, allocates GPU resources, executes training, evaluates checkpoints, iterates. Targets drug discovery, chemistry, biology, robotics — not consumer chatbots. First model/product promised in coming months.
No public safety framework disclosed as of June 2026.
Related
- automated-research
- anthropic
- ai-investment-trends
- venture-capital
- recursive-self-improvement
- mirendil-200m-ai-research-automation