What we’ve built, and what we haven’t.
Case studies with the architecture, the decisions behind it, and the parts that were genuinely difficult. A prototype is called a prototype.
What the labels mean
- Built
- Implemented and usable today.
- Prototype
- Core works end to end; not production-hardened.
- Concept
- Architected and specified. Not built.
Projects
- BuiltInfrastructure2026
AgentCost
AI Cost Observability Infrastructure
An SDK and backend that attribute LLM spend to the agent, run, step and tool that caused it.
- Python
- FastAPI
- PostgreSQL
- SDK Design
- +2
- PrototypeAI Systems2025
Spectra
Document Intelligence and Retrieval
A retrieval system over uploaded documents, with per-document scoping and streamed conversational answers.
- FastAPI
- ChromaDB
- Hugging Face
- Ollama
- +2
- PrototypeResearch2025
AeroLeaf
Satellite-Assisted Environmental Verification
An investigation into verifying reforestation claims from satellite imagery time series rather than site reports.
- Python
- NDVI
- Sentinel / Landsat
- React
- +2
- PrototypeFull Stack2025
Festival Creatives Platform
Dynamic Creative Generation at Scale
A template-driven platform that renders branded creative assets from structured data through a queued pipeline.
- TypeScript
- Turborepo
- Next.js
- PostgreSQL
- +2
- ConceptAutomation2026
Order Intelligence
Operational Intelligence for Fragmented Workflows
A system design for turning scattered operational messages into structured, trackable state — with humans confirming what the model extracts.
- LLM Extraction
- Human-in-the-Loop
- Workflow Systems
- System Design
Have a difficult problem worth building?
We’re interested in ambitious software, AI systems, research collaborations and real problems where careful engineering actually changes the outcome.