We are buying the coffee
What actually breaks when you change the model behind a production agent?
We are two engineers trying to answer that, and we need your help.
Boris (left) and Lohith (right).
Boris Goranov
Built the inference stack at a neocloud and moved production agents onto open-weight models. Before that, engineering at Google, Amazon and J.P. Morgan. Systems research at ETH Zurich, published at SOSP.
Lohith Yadala Chanchu
Published at ICML, ICLR and TMLR on getting models to check their own work. Built and sold the support agents now running on 20,000+ EV chargers. Before that, worked as an AI engineer/researcher at Qualcomm, JetBrains and Mercura (YC W25).
Supported by angels and scouts from OpenAI, Anthropic, Amazon, General Catalyst, a16z, and Lyceum.
What we are trying to figure out
- How did you end up on the model your agent runs on today, and has it changed since launch?
- Have you tried moving to a different one? What broke, and how did you find out it broke?
- What would you need to see before you let a different model answer a real customer?
We are not selling anything. There is no product, no demo, no deck. If we ever build something you would actually want, we will come back and ask you properly.
Who we are hoping to talk to
- You run an agent in production that talks to your customers
- Any stage, any size. A pre-seed team with 200 users is as useful to us as a Series C
- Extra useful if a model change has burned you before
Grab twenty minutes
Pick any slot that works. No prep needed.
Book a coffee chatPrefer email? lohith@orizon.fun