DevLand

Research

We study how machines see: how a camera becomes an account of what happened, where, and when, and how that account can be checked.

The work runs across six fields that only make sense together. Every result is measured on held‑out data before it is believed.

• Computer vision

  Detection, tracking and reconstruction from ordinary cameras: people, objects and motion recovered from footage never meant for measurement.

• Vision‑language models

  Models that describe what they see in plain language and point back to the frame they saw it in.

• Databases and retrieval

  Finding one moment across hours of video and millions of pages, with the path from question to source kept intact.

• Machine learning

  Training on measured data, gating on data the model has never seen, and keeping the record of both.

• Reinforcement learning

  Agents that learn where to look longer, which recording to open next, and when to stop and ask.

• Orchestration

  Many models in one pipeline: scheduling, verification and an audit trail that follows every output back to its input.