I spent my working life listening for that leak - building the systems that catch a pattern before it's obvious to anyone else.
Long before spreadsheets, I was drawn to the places where a pattern hides just under a surface of noise: weather, traffic, the rhythm of a market before it turns. It looked like curiosity. It was actually a search for the smallest unit of trust a person can reasonably place in a prediction.
That search became a discipline. Time series, then language, then whatever the problem in front of me needed modeled and defended in front of people who had never heard of a loss function. The tools changed - spreadsheets, then spiking neurons, then transformers reading a decade of documents in an afternoon. The question underneath never did: what is this data actually trying to tell us, and how much of it can we honestly stand behind?
Generative AI made it possible to process a decade of documents before lunch. It did not make it easier to know which of those documents actually mattered, or why a model reached the conclusion it did. Most teams solve for throughput and hope interpretability catches up later. I've never found that trade acceptable - not out of caution for its own sake, but because where the stakes are real, "it usually works" is a sentence nobody wants to defend later.
My response isn't slower systems. It's better-built ones: retrieval pipelines that can point back to their sources, forecasts that arrive with their own error bars, teams that ship an honest MVP without mortgaging the roadmap to get there. A model that can't explain itself hasn't finished being built.
Pick whichever lens is relevant to you.
I build systems that read faster than any team could and still know where every claim came from - retrieval-augmented pipelines and language models wired into workflows where a wrong answer has a real cost. Production, not demos.
Long before generative AI, I was modeling what happens next - demand curves, churn, anomalies buried in years of sensor data. I still think the best forecasting systems are the ones that admit their own uncertainty out loud, instead of hiding it behind a single number.
I've led engineering and data science teams from a rough idea to something a business actually depends on - translating between what a model can do and what a stakeholder is willing to trust. Roadmaps, not just repositories.
A doctorate in neural networks and time-series forecasting, most recently spiking neural networks - models that fire the way biological neurons do, sparingly, only when something is genuinely worth signaling. It's less about building a bigger model and more about asking how little a system can get away with knowing and still be right. I've taken that question outside my own field too, presenting it to a room of physicists at a statistical physics conference - what a machine can learn about the shape of short-term memory. The same instinct runs through everything else here: don't spend a signal on something that isn't there.
The best collaborations usually start with a question, not a pitch.