1
Workflow evaluation
How many hours do AI workflows really return in small businesses? Measured on anonymised project data, with client consent, and released as an open benchmark.
Serves Groundwork
Baseline · a bit175 company
Research that makes AI useful at work.
Four research tracks, each answering questions our clients and companies face every day. We publish our methods, our data where we can, and our negative results.
Our publication rule: we never publish anything that could identify a client, or findings that carry a security risk.
Research tracks
Research that serves real work.
2
West African language AI
Speech and text for Yoruba, Hausa, Igbo and Nigerian Pidgin, where commercial models are weakest.
Serves Commonplace
3
Efficient inference
Routing, caching and smaller models, so products stay affordable in price-sensitive markets.
Serves Commonplace
4
Agent reliability
How and why AI agents fail inside business workflows, and how to catch failures before clients do.
Serves GroundworkServes Kindling
Research with us.
We are looking for university groups, non-profits and companies to share data, compute or questions. Partners are credited on everything they contribute to.