Prototypes, proofs of concept and internal tools built by our founders to test an idea quickly. For production systems delivered for clients, see the case studies.
Runs object detection, labelling and bounding boxes across 9 vision models from 4 providers in parallel, scores each one for accuracy, completeness, hallucination and cost, and ranks them on a live leaderboard.
Result Best overall: GPT-5.4, 4.63 out of 5. Best value: Gemini 3.1 Flash-Lite (preview), 4.05 out of 5 at $0.0008 a run. A quality-versus-cost slider re-ranks the models instantly.
A fine-tuned model that writes the JSON configuration Creatomate needs to render an animated video, so a team can produce video variations without configuring every animation by hand.
Result Trained on hand-written and synthetic examples. A test harness checks JSON validity and template alignment before a fine-tuned version is used.
Watches Product Hunt launches, qualifies each company with AI, enriches founder context with Apollo and Tavily, matches it to relevant past work and drafts personalised outreach, saving the results to Airtable. Sending leads on to HubSpot stays a manual step, after review.
We can build a focused prototype to prove it works, then take it to production with the evaluation, monitoring and handover it needs.
