RunLog Atlas
Building the infrastructure layer for human-in-the-loop AI systems. Atlas makes review scale with ambiguity, not volume—enabling systems to get cheaper and more accurate over time.
Meta
Built ML systems serving 50M+ daily active users. Improved hate organization detection by 15%, directly impacting 11M+ profiles and reducing false negatives by 39%.
Key Insight
Learned that confidence scores without operational routing are meaningless. Systems need to act on what they know—and what they don't. This insight became foundational to RunLog Atlas.
Bridge
Built document intelligence pipelines processing 10,000+ documents daily. Best-in-class extraction. High confidence scores. Yet review scaled linearly—more docs meant more human hours.
The Bottleneck
Saw firsthand why extraction alone never scales. Deadlines broke teams, not models. This problem became the genesis of RunLog Atlas—making review scale with ambiguity, not volume.
Dirac
Head of AI. Set up systems from scratch handling 1M+ geometries. Led multiple 0→1 projects, implementing advanced algorithms that reduced user-facing latency by 70% and workflow interruptions by 90%.
Stax
Founded Stax, growing to 400+ weekly active users across 4 colleges, supporting 15,000+ classes with personalized recommendations. Invested $10K and managed the entire product lifecycle from ideation to launch.
Lessons Learned
Validated market needs, developed user-centric solutions, and drove rapid growth. While we ultimately pivoted, the experience provided invaluable lessons about product development, user acquisition, and market validation that inform RunLog AI today.
From Meta's scale to Bridge's insights to building RunLog Atlas—I'm focused on solving production AI challenges.