DigiCrafter is an AI studio working at the intersection of pragmatic engineering and human-centered design. We're small on purpose. We build things that actually run, in production, on infrastructure clients control.
Most of the noise about AI sorts into two camps. One says it will replace us. The other says it's a toy. Both are wrong — and wrong in the same way: they assume intelligence is the thing humans put into words.
It isn't. Most of what we know is in our hands, our nerves, the room we're standing in. Three principles follow from that.
Books, code, transcripts, conversations — digested and made accessible. That's the gift. But the judgment that runs a clinic, the taste that picks the right photo, the read on a room: still human. AI is a subset of human intelligence, not a successor to it.
So we build systems that hand intelligence to people. Not systems that try to replace them.
AI is non-deterministic. So are humans. So are great teams. You don't trust a new hire on day one; you trust them after six weeks of seeing how they handle the hard cases. AI is the same.
So we ship narrow proofs first, validate them against your actual operations, then expand only what's earned its place.
An AI can produce ten options in a second. Choosing the right one is a human act. Owning the consequence is a human responsibility. The human-in-the-loop is load-bearing, not a courtesy.
So we keep your team in the loop, always. We don't ghost-build. The system is yours, and you understand how it works.
AI samples; humans know. AI extends; humans decide.
If you've been sold AI three times this quarter and you're tired — we're probably your speed.
Every system we build is judged by what it does for the person using it — not by how impressive its architecture diagram looks. The user is the spec.
We build on open-weight models running on hardware our clients own. No vendor lock-in. No data leaving the client's network. Predictable cost, full control.
The smallest tool that can do the job, every time. We've replaced microservices with shell scripts and shipped Excel-driven workflows when Excel was the right answer.
We tell clients when AI isn't the answer. We tell them when our first plan was wrong. We tell them when a feature isn't working before they ask. Trust compounds.
For 20 years I've worked inside complex enterprise systems — sitting with CTOs, operations heads, and risk officers at tier-1 banks, federal agencies, and regulated firms. The role has been consistent: listen past the stated requirement to the real one, design a system that fits the actual world it has to run in, build it with the team, and stand by it through go-live and the years of support that follow. Twelve years at Hewlett Packard Enterprise as a senior consultant and architect. Eight at ITRS Group, ending as Manager, Managed Services — leading the engagement that supported Bank of America's global Geneos operations across three time zones. Perfect NPS scores from every client across that tenure, and an earlier six months as Interim Head of Professional Services for the Americas. Before that, five years building systems-level software at Computer Associates.
That background isn't decorative — it's the foundation for what DigiCrafter does now. The OTC trade confirmation pipelines I built for Credit Suisse, Bear Stearns, and Citibank taught me what "auditable, citable, reproducible" means when a workflow failure costs millions — the same discipline that now shapes how we build retrieval systems for legal and pharma. The Autonomy IDOL deployments at Scotiabank's KYC unit and Bonneville Power Administration's federal indexing were enterprise retrieval at corpus scale, before anyone called it RAG. The years supporting Geneos infrastructure for JPMorgan, Wells Fargo, ITG, and StateStreet taught me how production systems actually run — and fail — under real load, which is exactly what we now apply to verifying AI agents in production. Add Freddie Mac, KPMG (Global), Liberty Mutual, Kilpatrick, Stone Pigman, Carnival, CBS — fifty-plus engagements across worlds that don't usually talk to each other. Modern AI is the same craft, sharpened by a new tool.
In October 2025, I went all-in on AI. Not because the hype peaked — because the underlying tech finally got good enough to do what the last 20 years of enterprise software couldn't. That's where the leverage is.
DigiCrafter is the studio I built around that conviction. We work with regulated enterprises — legal, pharma, financial services, education — where AI has to be grounded, citable, and trustworthy enough to ship. In seven months we've put eleven initiatives in flight: legal-grade retrieval over the Indian indirect-tax corpus, a 12-year personal photo archive with on-device face clustering, agentic QA harnesses that verify themselves deterministically, and more. Two are in production.
The throughline across all of it: I show up, I listen, I build something that runs in production, and I earn trust over years. If you've been sold AI by people who haven't shipped enterprise software before — that's the difference. I also give free 45-minute career calls to anyone moving into AI, and write field notes about what we're learning. Drop a line at rg@digicrafter.ai or book a call.
A side project to wire one chat across 20+ messaging apps. The lesson: a single user with real needs is a better forcing function than a hypothetical enterprise customer.
Studio formally launched. First client work in pharma competitive intelligence, then legal AI. Built the rig because cloud bills got silly fast.
Took on the largest engagement to date: a quote-grounded retrieval system over the Indian tax-law corpus. Forced us to figure out structure-aware chunking and on-prem inference at scale.
Working with a handful of clients per quarter on projects that matter. Publishing field notes from each build so the lessons compound publicly, not just in private playbooks.
Thirty-minute call. No slides. No sales pitch. Just whether we can actually help.
Book a 30-min call