Systems that read things
Documents, claims, forms, reports, and instrument output turned into clean structured data your systems can actually use.
Forward deployed engineering
Some problems need an engineer in the room, not a vendor.
You have a workflow that eats your team’s week. Everyone knows what it is. It hasn’t been fixed because fixing it takes someone who will sit with your data, learn how your operation actually runs, and build against the real thing.
Fifteen years in operations where the data is difficult and the stakes are real
Data Science Program Manager, leading AI and machine-learning delivery for clients from problem scoping through production systems. Healthcare data carries regulatory weight: the reporting has to be right, and it has to be explainable to the people who are accountable for it.
Senior Data Scientist. Built machine-learning models on Advanced Metering Infrastructure data to predict power outages and equipment failures, using time-series forecasting and event-based anomaly detection. Led an outage-analytics system that improved reporting accuracy and cut the reliance on manual verification, plus the pipeline that unified utility datasets that had never been joined before.
Research Scientist on the X-ray CT explosive-detection systems the TSA uses in airport screening. Built the statistical models and machine-vision methods that evaluate whether those scanners are performing correctly, including how performance degrades over time. Lead inventor on two granted patents (US 11,430,109 B2 and US 11,734,816 B2). The analysis frameworks took evaluation from weeks to hours.
PhD research on the ALICE experiment at the Large Hadron Collider, with annual onsite shifts at CERN. Built a C++ framework to analyze femtoscopic correlations in lead-lead collisions and applied machine-learning methods to extract meaningful parameters from very large detector datasets. Author on three publications, including Phys. Rev. C 103, 055201.
The third option
Software companies sell a product. Consultants sell a recommendation.
A software company sells you a product and hopes your problem fits it. A consulting firm sells you a recommendation and leaves before anything is built. Both models assume your environment is normal. Yours isn’t. The data is messy, the systems are older than the vendors admit, and being wrong has consequences.
Forward deployed engineering is the third option: an engineer embedded in your operation, writing code against your actual systems, shipping working software while the engagement is still running.
How it works
Inside your environment, not adjacent to it.
01
Week one is the operation
Not requirements gathering. I spend the first week understanding how the work actually gets done, including the parts nobody documented.
02
Built against the real thing
Code written against your actual systems and your actual data, not a sanitized extract that behaves the way everyone wishes it did.
03
Yours when I leave
Something running in production, your team trained on how it works, the documentation in your hands. No dependency on me afterward.
No pilot that dies at the handoff. No proof of concept that needs a second budget to become real.
Why me
Most people selling this came from consumer software.
They have never worked in an environment where a bad deploy has a regulator, a patient, or a substation on the other end of it.
I’m Jesse Buxton, and Buxton Innovations is my firm. It is deliberately small: the person who scopes the work is the person who writes the code, and there is no bench to hand you off to.
I have spent fifteen years in operations where the data is genuinely difficult: experimental physics at CERN, critical infrastructure at Battelle, grid and utility analytics, and healthcare data systems. PhD in physics from Ohio State, MBA from Fisher, two patents. I can build the system and then explain it to your board without a translator in between.
That combination is the whole offer. The engineering alone is a commodity. The engineering plus the ability to operate inside a regulated, instrumented, politically real environment is not.
Typical work
Where this work usually lands.
Reporting that runs itself
The recurring report that costs someone a day a week, produced automatically and correctly.
Multi-step operational workflows
The process that touches four systems and three judgment calls, handled by something built to make those calls the way your best person would.
Analytics that survive real data
Models and pipelines built for operations where the inputs are incomplete, inconsistent, and occasionally wrong.
Engagements
Three ways to work together.
Embedded Sprint
One painful workflow, proven fast
Two weeks
One workflow running end to end in your environment.
Forward Deployed Engagement
Your core system, built right
Four to six weeks
A production system integrated into your stack, with handoff documentation and your team trained on it.
Embedded Retainer
Ongoing, evolving needs
Monthly
A dedicated engineer on standing capacity, new work as it comes up.
Pricing is scoped to the engagement. Most teams start with a sprint on the workflow costing the most time, then expand once it’s obviously working.
Get started
Let’s scope the problem.
Describe the workflow costing your team the most time. You’ll leave the first conversation with a scope, a timeline, and a number.
Or reach me directly at jesse@buxtin.net Buxton Innovations LLC · Ohio