model-optimization

(4)
Legend
Viraat Das
Founder · Engineer · Executive

Viraat Das

Viraat Das is the cofounder and CEO of Exla, a Y Combinator (Winter 2025) startup building AI infrastructure. A former Amazon machine learning engineer who finished his Purdue computer science degree in two and a half years, he first shipped an SDK that squeezes large transformer models onto edge hardware like NVIDIA Jetsons, then aimed the company at Libra, an always-on AI site reliability engineer that reads production logs, traces failures to the deploy that caused them, and opens fix pull requests while the team sleeps.

viraat-das · exlaRead →
Company
Exla
Ai · Developer Tools · Enterprise

Exla

Exla (YC W25) is a San Francisco startup building Libra, an AI site-reliability engineer that watches production logs, traces and deploys around the clock. When something breaks, Libra detects the failure, finds the deploy that caused it, opens a fix pull request with a regression test, and verifies recovery in production - so engineers do not have to wake up at 3am. The company began as an SDK for running large transformer models on edge devices and pivoted to autonomous incident response.

ai · sreRead →
Company
Osmosis
Ai · Developer Tools · Enterprise

Osmosis

Osmosis is a San Francisco reinforcement learning platform (YC W25) that helps companies fine-tune open-source models to beat foundation models on cost, latency, and reliability. Founded by Kasey Zhang and Andy Lyu, it lets teams train task-specific AI agents with multi-turn tool use over the Model Context Protocol, then export or self-host the weights they own.

reinforcement-learning · rl-fine-tuningRead →
Company
ModelCat AI
Ai · Hardware · Developer Tools

ModelCat AI

ModelCat AI, formerly the low-power chip pioneer Eta Compute, builds an agentic, hardware-aware platform that automatically designs, trains, optimizes and validates machine-learning models for embedded, edge and IoT devices. Its patented AI-in-the-Loop workflow, backed by a physical hardware farm that calibrates results against real silicon, compresses a model-development cycle that once took 12 to 24 months down to a matter of days. The Sunnyvale company works with chipmakers including NXP and Alif Semiconductor to hand developers ready-to-run models tuned to the exact constraints of the hardware they will run on.

edge-ai · tinymlRead →