Louis Collard
I'm a lifelong builder and occasional troublemaker. I’ve written bare-metal security firmware for Google, contributed to the Linux kernel, managed data pipelines responsible for billions in ad revenue, and navigated a venture-backed startup through a bet-the-company pivot to a successful M&A exit.Having burned through all the skepticism I could muster, I've come full circle – back to the neural nets I studied in my undergrad, before deep learning Transformed them from dead end to all-encompassing future.I'm currently working on:Into the Matrix – an open-ended, first-principles exploration of contemporary AI, starting with rule-following. I have yet to see a model that can adhere strictly to non-trivial policy, with compliance that doesn't degrade as context grows – or read a paper describing what one might look like. My bet here is that without that, an LLM session will never converge to a truly stable state, meaning chat sessions / agentic loops can only terminate at a locally plausible point – easily destabilized by any follow-up message. This bet carries through to the products below.Kalaido – an AI-native take on Obsidian and friends. Chat-with-your-notes apps automate the homework required to make a "traditional" PKM useful, but don't solve the real problem – inevitable staleness. Kalaido puts the emphasis on the lens that generated a document rather than the document itself, turning a static store into a live synthesis of your current reality. Snapshots and a diff-driven reconcile step hand you control over how it all converges.Agent Smith – a lightweight, all-in-one framework for building deterministic chat-based apps through persistent, explicitly mutable state. Two apps based on this in the queue.Spock – outlines first, coloring in second. A toolkit that makes a coding agent settle the wireframes, UX flows and domain model before it writes a line of code, so the arbitrary calls it makes on day one don't persist as decisions equivalent in weight to your own. Experimental; prototype works.I have previously consulted as an eng "fixer" / Fractional CTO, and still enjoy saving people the cost of a wild goose chase. If you need help with something adjacent to the projects above, I'd be happy to chat.