We build AI systems that go into production and stay there.
AI implementation for companies putting AI to work, research centers, and government agencies. Eighteen years of building things that had to keep working after we left.
Thirty minutes, no deck, and a straight answer on whether it is worth building. Or two weeks and a fixed fee if you have a pilot that has not shipped and want the answer in writing.
A typical first mission
8 weeks
- Week 0–2
-
Reconnaissance
We map the workflow, the data, and the review process that will decide whether this can go live. You get a written brief with the scope and the conditions under which we walk away.
- Week 2–6
-
Build
The system gets built against your real data, in your environment, with security and records people in the room from the first week rather than the last.
- Week 6–8
-
Handover
Your team is trained, the runbook is written, and the system runs without us. No retainer required to keep it alive.
Outcome: in production, owned by you
Eighteen years shipping software: research programs, state and city government, and commercial platforms built from scratch
- National Science Foundation
- DC Public Works
- State of Maryland
- Penn State
- Autodesk
- Booz Allen Hamilton
- Quorum
- Travel insurance
SOC 2 Type II · 2024
Delivered under emergency conditions, for state and city government
- 750K
- people vaccinated through the scheduling system we built for Maryland.
- 45M
- text messages sent and received for two state health departments during contact tracing.
- 1M
- vaccination appointments scheduled, at ten thousand a day when it peaked.
- 9 years
- continuous delivery for a city public works department, without a gap.
01
What we build
Whatever we build, it has to run without us.
That is the one thing the three lanes below have in common. Systems you own outright, agents that take work off your people, and two products we host and run. Different contracts, different risks, one definition of done: it is in production, it survived review, and your own team can operate it.
-
Systems
Built once, yours to keep
The data spine under a department, the platform under a business, the integrations between systems that were never meant to talk. Scoped against your real data, then handed over with a runbook and a trained team. Eighteen years of these, from an enterprise service bus to a pandemic response to a commerce platform still selling today.
Federal · state · local · commercial
What we have shipped
-
Agents
Work taken off people, not renamed
Expert people spending their week on data entry, chasing and reformatting: that is the pattern we look for. The agent takes that part, shows its work, and asks before anything that matters. In production today: a report that assembles itself, and an advisor grounded in a program’s own documents. Plus the dozen agents that run this firm, which is where we learned what breaks.
Any operation where expert work turned into data entry
Agents in production
-
Products
Licensed, and we run them
Software that already exists, hosted and maintained by us, so you buy a running system rather than a project. Two of them today: SimplyScholar for research centers, and a service alerting platform for city public works. Both in production for years rather than months, priced as an annual line item.
NSF centers · city departments
See both products
02
The problem
Why do AI pilots stall?
Almost never because the model was not good enough. The demo usually works. It is everything after the demo that kills it, and it is the same three things almost every time.
95%
of enterprise AI pilots produce no measurable impact on the bottom line.
- It was never built to pass review
- Security, privacy, and accessibility turn up after the demo lands. The rebuild costs more than the build did, so it does not happen.
- Nobody could answer for the output
- A confident wrong answer is worse than no answer, and in anything that gets checked it is worse still. If a user cannot check the result against a source in one click, the system does not get trusted with real work.
- The people who built it left
- A system nobody on staff can change is a system that quietly stops being used. Handover is a deliverable, not a courtesy at the end.
03
Where most people start
The Production Readiness Review
Two weeks. You get a written answer to the only question that matters: can this thing actually go live, and if not, what is stopping it.
- Who it is for
- You have a pilot that works in the demo and has not shipped. Or you have one about to go into security review and you would rather find out now than in the meeting.
- What you get
- A written verdict of ship, fix, or stop. The specific things that will fail review, named, with what each one takes to close. What finishing it would cost and how long it would take. Whoever owns the budget can read it without a translator.
- The fee is not rebated against a build
- Firms that credit the audit against the follow-on project have a reason to recommend the project. We do not do that. If the honest answer is stop, we write stop, and you have saved considerably more than the fee.
Engagement spec
- Duration
- Two weeks
- Fee
- Fixed, quoted before we start
- Deliverable
- Written review and verdict
- Your time
- About four hours
- Obligation
- None
We reply within two business days.
04
Selected work
Systems that are still running.
-
Two states through a pandemic, as prime contractor
Contact tracing, mass vaccination scheduling, community testing, and the letters people showed their employers. 45 million messages across both states, 1.5 million contact tracing cases, and around 750,000 people vaccinated through the system we built.
Maryland and Delaware
Prime contractor, 2020–2023 -
A state website that stayed up at 150,000 requests a minute
Rebuilt and hardened in four working days, ahead of a governor's press conference, because the alternative was the site falling over live on television.
State of Maryland
Four days -
A hundred million readings off a snowplow fleet, answerable in real time
Weekly reporting down to fifteen minutes, across five to seven core systems. A dozen agency systems joined into one data warehouse, GIS underneath, real-time entry from crews in the field. Ten years, four administrations and a pandemic, without a gap.
DC Department of Public Works
Continuous since July 2016 -
$120 million a year in policies, across eight countries
A travel insurance commerce platform we architected and built from nothing: eight countries on three continents, five currencies, fourteen languages, thirty-odd sites, three AWS regions. We finished in 2015. It is still selling policies today.
Travel insurance group
Subcontract, 2012–2015 -
A failing student records system, taken over and finished
Case management for a public school system, inherited mid-crisis. We got it to a first release, then led four more years of development on it. All of it under FERPA.
Public school system
Subcontract, 2010–2014 -
Four weeks of annual reporting, down to a two-day review
The report now assembles itself from data the center already has. The coordinator reviews and edits instead of chasing fifty researchers for it.
NSF research center
In production -
Proposal help that knows the solicitation better than the deadline allows
Grounded in the program guidance and the office's own funded proposals, so the advice is specific enough to act on at four in the afternoon.
University research office
In production
05
The firm
Who actually does the work?
A senior practitioner on every engagement, start to finish. No bench, and no handoff to a delivery team you have not met. The person who scopes your work is the person who builds it, and the same person is in the room when it goes to review.
- We were our own first client
- Roughly a dozen agents run this firm day to day: scheduled briefings, invoice reconciliation, meeting notes, pipeline research, inbox triage, and internal search across everything we know. We are not recommending a way of working that we have not lived with.
- Eighteen years before the AI part
- A decade inside NSF research programs, ten years and counting with a city public works department, prime contractor on two state pandemic responses. And on the commercial side, a travel insurance platform we architected from nothing that still sells in eight countries, plus two startups: one whose product we built from scratch and took through its own SOC 2, one we advised through the growth that broke its architecture.
- We have sat on both sides of the audit
- We have been through SOC 2 Type II ourselves, and this year we took a client through their SOC 2 Type 2: the penetration testing, and the evidence responses to the auditors on operational, configuration, and vendor controls. They got their report. We have built and run HIPAA infrastructure holding live patient records for a state health department, and FERPA-covered case management for a public school system. We build everything to that bar, including for clients who have no auditor to answer to, because it is the difference between a demo and a system.
The oldest thing we built is sixteen years old.
A case management system for a public school system, built in 2010 and running under FERPA, still in service as far as we know. The one after it, a travel insurance commerce platform from 2012, is definitely still selling: eight countries, five currencies, three continents. And the oldest system we still operate ourselves is a decade-old service bus for a city department. Anyone can ship something that demos. Keeping it alive while every administration, dependency and vendor underneath it changes is a different skill, and it is the one this firm is built around.
06
Writing
What we are working on
Notes from the builds. What worked, what we got wrong, and what we would do differently.
-
July 2026 · 14 min
Papa Claude and Baby Claude: put the expensive model at design time
An agent that worked on day one and would have bankrupted a filing season. The fix was an architecture, not a prompt. Open source.
-
August 2026 · 12 min
The annual report is the best AI project your center is not doing
Four weeks of coordinator time, every year, in every center. Nobody writes about it because it is not glamorous.
-
August 2026 · 10 min
How we run this firm on a dozen agents
What we automated, what we tried to automate and reversed, and what it cost to find out.
07
Common questions
- What happens in a Production Readiness Review?
- Two weeks. We read the code, talk to the people who built it and the people who have to approve it, and put your pilot through the review it will eventually face. You get a written verdict of ship, fix, or stop, with the blocking issues named and costed. It takes about four hours of your team's time and carries no obligation to do anything afterwards.
- How long does a build take?
- A first mission runs six to eight weeks from brief to a working system in your environment. Larger builds run three to six months. We do not run open-ended engagements, and we do not need a retainer to keep what we built alive.
- Can you fix an AI pilot we already started?
- Often, yes. A stalled pilot usually fails on data access, review readiness, or ownership rather than on the model. We start by finding out which one it is, and we will tell you if the honest answer is to stop.
- Do we have to be a research center or a government agency?
- No. We work with any organization trying to get AI into real use, and plenty of that work has nothing attached to it at all. Where a project does carry HIPAA or a state security review, that is the same build under harder conditions. You get the same standard either way.
- Can you work with our regulated data?
- We have designed and run HIPAA-compliant infrastructure holding live patient records for a state health department, on hardened AWS with unified security logging and single sign-on, and passed the state's security review to do it. We have also run FERPA-covered case management for a public school system. If your data carries a regime we have not worked under, we will say so on the first call rather than discover it during the build.
- Are you SOC 2 certified?
- We hold a SOC 2 Type II report from Prescient Assurance dated 2024. We renew when an engagement requires it rather than on a calendar, so no, we have not been audited in the last two years, and we are happy to send you the 2024 report. What did not lapse is the practice: we run HIPAA-compliant infrastructure holding live patient records today, hardened to CIS Level 1, and earlier this year we took a client through their SOC 2 Type 2, running the penetration testing and answering the auditors directly on operational, configuration, and vendor-assessment controls. They passed. If a current attestation is a hard requirement for your procurement, say so on the first call and we will tell you straight away whether we can meet it and what renewal would take.
- Can you be bought on a contract vehicle we already have?
- We hold Maryland CATS+ across all ten functional areas and we are registered in SAM.gov, both current. For a research center the work is usually funded straight from the grant, which avoids institutional procurement entirely. On SOC 2, see the answer above.
- How big is your team?
- Small, deliberately. Three senior people, a bench of fractional senior specialists we bring in per mission, and the agent stack that runs our own operations. What we do not do is keep juniors on a bench and find them something to do on your budget. That is why every mission is scoped against one workflow, priced as a fixed fee, and has a defined end. If your work genuinely needs a standing delivery team of ten, we are the wrong firm, and we will tell you on the first call rather than take it and hope.
- Do you write AI strategy documents?
- No. We build the systems a strategy document would describe. If a roadmap is genuinely what you need, we are the wrong firm and we will say so on the first call.
- What does it cost?
- The review is a fixed fee, quoted before it starts, and it is not credited against later work. Builds are fixed-fee and scoped against one workflow, so you know the number before anyone starts. We will give you a range on the first call rather than after three meetings.
Bring us one workflow that is costing you time.
Thirty minutes, no deck. You will leave the call knowing whether it is worth building, roughly what it would cost, and whether we are the right people to do it.