Hello!

I’m Ajey Pandey.

I’m an amateur road cyclist, an amateur photographer, a washed-up musician, a recent student of late Soviet history, a “friend of the show” for the Foundation for American Innovation, and a Senior Analyst (a Professional Knower of Things) for Halcyon.

I’m an engineer by training, a civil servant from experience, a technical writer by natural talent, and an insufferable art boy in person.

This is my personal website. It has served many purposes, but my current output is elsewhere.

For my resume, go here.

For my writing at my previous employer, SemiAnalysis, go here.

What Do I Do?

Halcyon is an AI-enabled data platform for the written record of electric power regulation.

If you’re familiar with the websites for state public utility commissions (PUCs), the quasi-judicial regulatory bodies that adjudicate the regulation of electric utilities in the United States, you’ll know they all suck. Each PUC operates differently; each one has a bespoke terminology and case law corpus; and none of them made their documents easily accessible, much less searchable.

If you’re savvy with agentic large language models like Anthropic’s Opus 4.6+ or OpenAI’s GPT 5.5+, you might think you have your answer: just throw tokens at the problem, either by making the agents crawl around in “deep research” workflows, or by vibe-coding scrapers to pull documents, or else by hand-downloading a stack of PDFs and making the model sort out what the hell is going on.

This is, of course, not as simple as you think. We may already have artificial general intelligences accessible by API, but they’re far from perfect. They don’t have the tacit knowledge of a professional ISO-watcher like yours truly, and they’re limited by both context window (how much information an AI instance can hold) and by the information that makes it into that context window—by the way, downloading a PUC’s document corpus is not trivial. The websites fight back.

Halcyon addresses this problem at both a technical and a subject-matter-expert (SME) level. The technology stack is, to over-simplify, a harness for a large language model: a scan-chunk-retrieve stack that allows for a smaller (cheaper) AI model and fewer mistakes. And the SME sauce points this technical stack at problems relevant to the people staring at the written record of electric service.

I was a power user of the Halcyon toolset before I started working here. My job is to stay a power user, push the toolset to its limits, predict the questions customers might ask of the system, and sketch out the shape of the likely answer.

I remain a generalist analyst, a Professional Knower of Things on Electric Service. But my deliverable is no longer an 8,000-word report: it’s now a product revision spec so that a Halcyon customer can write that report faster.

My job, ultimately, is to productize my knowledge, curiosity, and intuition of other analysts’ minds.

My job is to make you a better utility analyst.