Helping companies use their data to build better AI models

Talc builds datasets 100x faster than human labels while outperforming humans in quality. We build custom datasets in any domain, and have replaced human labels for teams in fields ranging from Medical care to Insurance.

Team Size:3
Location:Seattle, WA
Group Partner:Michael Seibel

Active Founders

Matt Lee

Ex-Airbnb, Ex-FB SWE. At Facebook I worked on Election Integrity, using graph algorithms to detect coordinated activity, from foreign election interference to terrorist propaganda. At Airbnb I worked on moonshots that will (maybe) ship late 2023. When I'm not working you can catch me snowboarding or climbing!

Matt Lee
Matt Lee

Max Kerr

Formerly doing privacy at FB and security at Microsoft. I was the tech lead for a 30 person engineering team at Facebook, and was involved in the privacy side of most major product launches over the last two years. I founded Talc to help people turn AI tech demos into production apps, along with my long time friend and climbing partner Matt.

Max Kerr
Max Kerr

Company Launches


We make it easy to debug and test LLM systems. Understand your AI by replaying sessions line by line and setting rules for it to follow moving forward.

Hey Everyone! We’re Matt & Max, the team behind Talc.AI

Matt: I’ve spent the last 6 years engineering at Facebook and Airbnb. At FB I worked on Election Integrity, developing new ways to detect foreign interference campaigns. Hit me up if you want to hear some war stories!

Max: I found my passion for tech early, teaching myself to code when I was 13. This journey most recently brought me to being the technical lead for a 30 person org at Facebook that reviewed every major product launch. I love the challenge of trying to understand complex systems and want to help people do the same with Talc.

We’ve been friends for almost a decade and couldn’t be more excited to be doing this together.

The Problem

Developing real LLM systems is still hard. Prompt engineering is full of trial and error, and debugging your chain and regression testing is a painful manual process.

The Solution

Replay any session your AI has had and tweak it until it's perfect. No more copy and pasting into a playground – re-run your sessions exactly as they happened and deeply understand their possible outputs. Have a bug on message 20? Hop into that context immediately and start fixing it.

Left: Walk through your AI message logs

Right: Edit the inputs and prompts that went into each message, all auto-populated for you

Test changes against your saved use cases for regression testing

What we’re asking

  • Is LLM debugging & testing holding you back? Reach out to us at talc.ai or book time directly on our calendars!
  • Do you know anyone working on LLMs in production? We’d love to learn from them! Help us get connected.
  • Share us with anyone who spends a lot of time debugging LLMs
    • Blurb: Talc.ai launched an LLM debugger so you don’t have to copy context into openai playground or jupyter notebooks any more. Reach out via founders@talc.ai.