Services
Bespoke AI software for evidence-heavy work.
I design and build custom software for professional firms — focused tools that use AI where it genuinely helps, with clear points where a person reviews and decides.
Thomas Demel, founder
I have spent more than a decade inside evidence-heavy professional work in R&D and innovation, after starting out as a software engineer. I know what a reviewer needs to see before they rely on something — and I know how to build, not just advise.
The principles behind Osini’s own products apply here too: evidence you can inspect, judgement that stays with your people, and data handled with care.
How an engagement works
Three stages, each agreed before it starts
01
Workflow assessment
I look at the current process, the outcome you want, the data involved and the constraints on it, and the build-versus-buy options. Sometimes the right answer is an existing tool, or no AI at all; if so, I will say so. You come away with a recommendation and, where it makes sense, a proposal for the next stage.
02
Scoped implementation
One agreed workflow, which I build and evaluate against examples you recognise. Outputs link back to their source material where that matters, and the points where a person reviews and decides are designed in from the start.
03
Handover and improvement
Documentation and training so your team understands what it has and how to run it. Ongoing support and further improvement are available, agreed separately rather than assumed.
Where bespoke earns its place
When a general-purpose tool isn’t enough
General-purpose assistants handle general tasks well. Bespoke software is worth building when the work makes more specific demands. These are illustrations, not descriptions of delivered client projects.
- Your firm's method, applied consistently
- A workflow that follows your documented criteria and checklists on every file, and routes the judgement calls to the right reviewer — for example, a first-pass review that flags what your senior people would look for, and why.
- Work that has to stand up to scrutiny
- Drafts and analyses where key statements link to their supporting material, checks and approvals are recorded, and a named person signs off — built to be inspected by a client, a regulator or a second reviewer.
- Knowledge that lives in the connections
- Material spread across systems and years, joined up — people, matters, documents and decisions, and how they relate — so questions can be answered across many matters, not just within one document. Often a knowledge graph; always chosen for the problem, not the vendor.
- Quality you can measure
- Tested against examples your experts recognise, so you can see how well it performs, and where it falls short, before relying on it — and notice if that changes.
- Data with strict handling rules
- Designed around where confidential material is stored, who can see it and what is logged — when a client, a regulator or your own policy sets the rules.
Scope
Is this a fit?
- A good fit
- Work with the demands above, where an off-the-shelf tool would need constant workarounds — and AI used where it genuinely helps rather than everywhere.
- Not a fit
- Configuring or rolling out off-the-shelf platforms. If you need Microsoft 365, Teams or Copilot set up, a specialist in those tools will serve you better than I would. Building bespoke tools that work alongside your existing Microsoft environment is a different matter — that is very much in scope.
Ways to work together
Projects, partnerships and mentoring
- A standard engagement
- An agreed scope and fee, with payment terms agreed before work begins.
- A co-development partnership
- For selected projects with potential beyond one firm, I may agree a co-development arrangement: a reduced development fee combined with agreed commercial rights or shared returns — preferential terms, a share of revenue, or referrals. Scope, ownership, reuse rights, confidentiality and ongoing responsibilities are agreed in writing before work begins.
- Technical mentoring
- A limited amount, for founders and teams building AI products — architecture, evaluation, and getting from a promising prototype to something you can rely on. Technical only: not legal, tax or financial advice.
Start a conversation
Discuss your workflow
A short email is the best start. Describe the process in a few sentences: what goes in, what comes out, who reviews it, and where the time or friction is.
Please do not send confidential records, client files or sensitive data at this stage. A description of the process is enough to tell whether it is worth a proper conversation.