SF AI Engineers: April

The April edition of SF AI Engineers brought together a sharp crowd of software developers, data scientists, and technical founders at House of AI in San Francisco. Three engineers got into the real stuff: visual memory systems, infrastructure rewrites, agent evaluation, and what it takes to train models at scale. If you missed it, the recordings are worth your time.

Full Event Recap

Watch the full session from start to finish, including all four talks and the Q&A discussions that followed each one.

Speaker Breakdowns

Wassim Gharbi, Founder, CoreViz

Wassim walked through how CoreViz is building a visual memory layer for AI, essentially a smarter way to store, tag, search, and edit media files. He made a strong case for why existing tools have not kept up with the volume of images and video that professionals deal with every day. The demo showed what it looks like when AI actually understands the content of a file, not just its name.

Tomasz Tunguz, General Partner, Theory Ventures

Tomasz shared how Theory Ventures migrated a core system from Ruby to Rust and came out the other side with 10x performance gains and roughly half the lines of code. It was a practical look at when a full rewrite is actually worth it and what the tradeoffs look like in practice. Less theory, more shipping.

Yury Markovsky, Director of AI, Prophecy

Yury made the case that evaluations and product requirements documents are not two separate things but the same thing. He walked through how Prophecy thinks about evals for autonomous agents serving data analysts, engineers, and business teams, and why getting this right early changes how you build. It reframed how most engineers in the room were thinking about testing.