What breaks when AI agents move past the prototype.

We are an independent engineering and research practice building open-source tools and reproducible research, prompted by the problems that appear once systems leave the demo: state, validation, retrieval, cost.

LONDON

Open Source

Things we’re working on

Tools and experiments prompted by problems we’ve seen when AI systems move beyond the prototype. Released work is open source; unfinished work is still taking shape.

Useful, over time

Private product

AskZoe

A private product exploring how AI can become more useful over time.

PERSONAL AIMEMORYRESEARCH

released

Felix

An agent harness that runs autonomous coding loops against real codebases without letting tests, types, or builds slide.

AGENT HARNESSQUALITY GATESCLI

released

Latch

A runtime that keeps structured conversations on track.

PROCESS RUNTIMECONVERSATIONAL STATETYPESCRIPT

released

RAGdoll

A RAG SDK for pipelines that have outgrown a notebook but do not need a managed retrieval platform.

RAGRETRIEVALKNOWLEDGE GRAPHS

released

Chomp

Smart URL and HTML parsers that convert web content into clean, LLM-friendly Markdown.

INGESTIONHTML TO MARKDOWNPYTHON

released

langchain-markitdown

LangChain document loaders built on the MarkItDown converter.

LANGCHAINDOCUMENT LOADERSPYPI

experimental

Echo

Fine-tuning transcription models for edge language cases.

WHISPERFINE-TUNINGNOTEBOOKS

Writing

Notes and research

Longer investigations into questions we find interesting, with enough detail for others to inspect or build on the work.

October 2025OSF, open access

The Optimistic Speculation Strategy: A Three-Layer Approach to Reducing Conversational AI Latency

Conversational systems wait for end-of-speech before they start working. OSS starts earlier: when a user pauses mid-sentence, it launches a speculative call on the partial transcript and keeps the result invisible, so the model is already warm by the time the user actually stops. Nothing speculative is ever shown, so a wrong guess gets discarded rather than corrected on screen.

Read the paper

Conversational AISpeculative decodingLatency reductionSpeech recognitionPredictive inferenceTurn-taking