AI and automation that make complicated processes simpler.
Practical applications and working products, not AI for its own sake. I build assistants, search and automations that plug into the systems you already use and save real time.
Start from a process that costs time or money, not from a model, and agree what success looks like.
Build a working prototype with your real data in days, so you can judge it on results.
Connect it to your phone lines, inboxes, databases and tools, so it fits how your team works.
Add evaluation, guardrails, cost control and monitoring before it handles real customers.
A cognitive architecture with needs, moods, brain chemistry, memory and goals running as code, and a language model as its inner voice.
Every morning a headless browser reads hundreds of Reddit complaints, an AI shortlists the real opportunities, researches the competition, and emails a digest.
Four AI agents and one orchestrator take a ticket from triage to code, QA, pull request and review, in a couple of minutes.
An experiment in handing an app's business logic, UI generation and orchestration to a language model. It worked better than it had any right to.
A vision model looks at who's walking past a digital billboard, then picks the ads most likely to land. Online-style targeting, out on the street.
From a ChatGPT-powered Siri Shortcut to a real-time voice assistant that runs a whole HomeKit house, reads the calendar and checks the cameras.
A chat assistant that reads OBD fault codes, searches a technical knowledge base by meaning, and explains car problems in plain English.
A GPT-4 ordering assistant that handles messy, mind-changing customers and outputs clean, structured orders.
A voice agent that took recruiter calls, pitched skills, qualified roles, politely shut down sales calls and saw straight through scams.
A pre-screening interview chat and a bulk CV processor that sorts hundreds of applications in minutes, with a written reason for every decision.
I fine-tuned GPT-3 on dialogue from the film Her to see how far training data alone can shape a model's voice.
A desk plant with real sensors, face recognition, voice and memory. Cloud AI services stitched together into one character.