Services
Prototype Sprint
- Duration
- 2 to 4 weeks
- My role
- I design and build it hands-on, in the open, in your repository.
A sales demo shows the happy path. A prototype sprint builds the capability on your real data, measures it, and tells you what production hardening remains and what it costs. You leave with something that runs in an environment you control, a baseline you can defend, and a decision you can make.
Discuss this engagement
Signs this is the right engagement
- Stakeholders are split on whether AI can do this job at all
- A vendor demo looked great and nobody trusts it
- You need a credible number, not a hunch, before committing a team
- The capability touches documents, transactions, or support cases where mistakes are costly
How it runs
- Week 0: scope one capability, agree the success metric, get data access
- Week 1: baseline pipeline running on your data with the evaluation harness in place
- Weeks 2 to 3: iterate on accuracy, add verification and escalation paths, instrument cost and latency
- Final week: hardening roadmap, handover, and a review call with the decision owner
What you get
- Working prototype on your data, deployed to an environment you control
- Evaluation harness with baseline metrics for accuracy, latency, and cost
- Production-hardening roadmap: what is missing and what it costs to add
- Weekly written progress notes and a final review call
What I need from you
- Representative sample data and access to the systems it touches
- A decision owner who can answer questions within a day
- A 30-minute check-in each week
When this is the wrong engagement
- You need a polished end-user product in the same four weeks
- The data cannot be shared in any form, even anonymized or synthetic
Afterwards
You own the code and the eval harness. If it proves out, I can stay on to harden it or hand it to your team.
Relevant work
Fintech, insurance and annuities · In production
Processing time cut from 4 hours to 8 minutes per application
A four-stage vision pipeline that reads handwritten annuity applications, cutting processing time from 4 hours to 8 minutes, with a monitored optimization loop gated by held-out evaluation.
Fintech, SME accounting · In production
100K transactions processed, over 90% automated end to end
An AI-native accounting platform that has matched, categorized, and posted 100K transactions across 10+ businesses with over 90% automated end to end, escalating only what it cannot resolve.
Related writing
Essay · September 2, 2026 · 12 min read
Why the most important design decision in a production AI system is deciding when the model is not allowed to answer, and how to engineer that refusal.
July 30, 2026 · 6 min read
Reading the Kimi K3 technical report from the seat of someone who builds agentic systems that have to run on Monday morning: the decisions were made by kernels, caches, and harnesses, not loss curves.
Pricing
Every engagement is a fixed quote agreed in writing before work starts. The quote moves on four things:
- How many systems, data sources, and environments are in scope
- Whether the work happens inside regulated or restricted environments (data residency, audit trails, access approvals)
- How fast you need it: a compressed timeline costs more than a steady one
- How much evaluation data and system access already exists versus needs to be built first
Start with a short note
Describe the system and where it hurts. I reply within two business days; if there is a fit, we schedule a 30-minute call and I send a written scope within a week.