A Claude Code skill that hands the assistant a whole persona, an accent, an attitude, and an energy that all correspond, and holds it in character for the entire session while the engineering stays senior grade. The code, comments, and commits stay clean, only the conversation takes on the voice. It ships with five deeply rooted voices built as love letters rather than caricatures, an invent-your-own mode, and a rule set for staying human instead of sounding like an AI in a costume. Open source, one Markdown file, no dependencies.
An autonomous quant trading platform built on a strict separation between thinking and acting. Claude, the non deterministic brain, writes strategies as content hashed data artifacts rather than runnable code, and a deterministic engine trades only the frozen artifact, so identical inputs always produce identical orders and the language model never sits in the execution path. Nothing reaches the market until it clears a walk forward validation gate that scores every candidate purely out of sample and raises its own bar as more ideas are tried, so overfit backtests get rejected instead of shipped. The whole thing runs around the clock as a hardened systemd service on a private server, paper trading equities and crypto, while a second Claude instance living on the same box narrates each event in plain language across a Telegram control plane.
A pocket 60.5 GHz radar built to measure in microns. It reads the phase of the returning wave, where one degree is 6.9 micrometres of movement, and does two things with it: count a heartbeat across still air, and recover a sound from a surface it never touches. Two firmwares across two processors share one protocol; the radar streams raw samples, the nRF turns phase into microns and runs each mode, with the signal processing validated in software before any hardware exists. The schematic and board are generated by Python and pass their checks with zero errors. Named the way Yoruba carries a borrowed word, the way radio becomes rédíò. My first hardware project, honest about the one measurement the whole design still hinges on.
A tiny language model, run from scratch in C. A complete transformer inference engine in a few hundred lines with no dependencies, no PyTorch, no NumPy, no BLAS, where the whole forward pass is written by hand: RMSNorm, rotary embeddings, grouped-query attention with a KV cache, SwiGLU, and sampling. It loads a pretrained model and generates coherent text; the goal is to shrink it until it runs on a microcontroller. Open source, with a live commit history.
A framework for modelling the complexity and uncertainty of human emotion. It takes an emotional history — a relationship, a friendship, a season of your own life — and lays it out as an interactive graph of observations, hypotheses, contradictions and lessons, built around one discipline: reality is not the model of reality. Every claim carries an honest evidence label, competing explanations coexist without a winner, contradictions are held open instead of resolved, and it never states another person's inner life as fact. Named from the Yoruba Oládìméjì — a thing that only exists between two people. Built with Next.js, React Flow and a d3-force layout.
An AI ghostwriting studio for apostolic and Spirit-filled ministry authors. A structured interview captures an author's theological voice into an editable "Voice DNA" profile, then the engine drafts full manuscript chapters in their exact style — every claim provenance-verified against their own words, with an in-studio AI assistant that writes as them, not for them.
A decision layer that sits on top of your product analytics. Connect PostHog or the Folio SDK, describe a growth goal in plain English, and Folio drafts the in-app nudges — banner, modal, and dashboard copy with CTAs — that you approve before they ship, then serves them through a headless decisions API with serve→click→convert attribution on every recommendation.
A shipped, premium AI Qur'an companion. A SwiftUI iOS app on a FastAPI/Postgres backend with Claude-powered reflections behind a safety + eval layer, geolocated prayer times, multi-reciter recitation audio, subscriptions, home-screen widgets, and an automated, personalized retention/notifications engine.
A self-hosted LLM pretraining pipeline built from scratch — data prep, training, and evaluation for small language models, owned end to end rather than calling someone else's API. The deep end of the stack.
An AI public-speaking coach with real-time conversational voice agents (ElevenLabs) and an LLM feedback pass that scores delivery and coaches each rehearsal. Credit-based billing, Google auth, the works.
Agentic prior-authorization for healthcare payers: LLM agents verify medical necessity with paragraph-level citations, gated by a deterministic citation verifier that blocks fabricated quotes outright — agents you can actually trust.
An AI editing studio that turns a long recording into a cut master plus captioned vertical clips — Claude tool-use proposes the cut list, ffmpeg renders it. Background jobs, transcription, and storage wired end to end.
An "aesthetic" AI recipe app where a single codebase ships as six niche App Store apps via an env switch. SwiftUI (WidgetKit, Live Activities, StoreKit) on a FastAPI/Postgres backend, with Claude recipe generation and fridge-photo vision.
A multi-agent AI platform — a Next.js studio over a distributed microservice backend (API gateway, agent engine, context engine), with Postgres and Qdrant powering long-term vector memory.
Sole engineer, across the whole stack, on a self-hosted multi-agent AI platform: an orchestrator agent routing to specialists (autonomous software engineering, institutional memory, generation, end-user assistants), a ~100-skill tool layer with self-extending tools and real-world integrations, self-hosted 14B LLM serving tuned for reliable tool-use, and a generative image/video engine running two 12B models on a single 12 GB GPU. From CUDA-level model serving up to the web app the client operates.
Applied LLMs
Modeling & rigor
Services & data
Ship & operate
Loading research papers...