Boston AI Developer
I'm a Boston-based AI developer and applied AI engineer. I don't build demos. I build AI into products and workflows that people use every day, and I spend a lot of time figuring out new ways to work with AI that most teams haven't tried yet. Here's some of what that looks like.
Using AI to research millions of historical records
I built an AI research pipeline that searched millions of handwritten records from 400 years of historical archives. In a few days, on a single GPU, it surfaced a forgotten meteorite report, three lost rhinos, and volcanic eruptions that were never recorded. It searches by meaning rather than keywords, and a small, fast decision model screens every passage so a larger model only reads the few worth a closer look. That screening step did the job of a researcher reading candidates one by one, at a fraction of the cost. I wrote up the full investigation and open-sourced the workflow as Antiquity.
Running a fleet of coding agents in parallel
One coding agent makes a developer roughly 10x faster. I've been working on what comes after that: running many agents at once, each in its own isolated copy of the codebase, while I review and merge their work. I gave a talk on multiplexing coding agents with Conductor and published every command and config file from it.
Along the way I've found that small, unglamorous habits matter most. A well-placed code comment is one of the most reliable ways to steer an agent, a good CLAUDE.md file saves hours, and MCP servers give agents the tools they need to finish the job.
Giving AI agents access to real systems
Most AI integrations stop at a chat box. I build agents that do things: send emails, query databases, book meetings, and hand work off to other agents. I wrote about agent orchestration on Rails, and released Agent Gateway, a Rails gem that lets an AI agent safely ask your production app how things are going. This website even has tools that AI agents can call through WebMCP.
RAG systems that answer from your own documents
When an organization's knowledge is scattered across PDFs, websites, and decades of policy documents, retrieval-augmented generation turns it into something people can just ask. I built one for a 150-year-old hiking club so its volunteers get plain-language answers with links back to the source. Here's how I built RAG on Ruby on Rails.
Fine-tuning custom AI models
Sometimes a prompt isn't enough and a model needs to behave in a very specific, consistent way. I've taken open-weight models and fine-tuned them end to end, and I spoke about it at Boston AI Week. Read how to create a custom AI model with fine-tuning.
Context engineering for AI features
The difference between an AI feature that works and one that doesn't is usually what's in the model's context, not the wording of the prompt. I wrote a guide to practical context engineering for teams shipping AI inside web and mobile apps.
Experiments
I also build things just to see what's possible, like a generative UI web browser, a voice agent that knows what I'm listening to on Spotify, and Mac apps that run Apple's on-device AI models.
Hire a Boston AI developer
I do this work for clients through JTPCK, my Boston AI consulting firm. It offers AI consulting for Boston businesses, AI integrations, RAG AI systems, and AI fine-tuning, alongside website development and mobile app development. You can see JTPCK's AI consulting work or read the AI archive search case study. If you'd like to talk about an AI project, send me an email.
