Services
Services
Three offers. Each begins with an assessment and ends with something you can run, test and read. We do not quote fixed results before seeing the code or the process.
S03 · AI Code Rescue & Technical Delivery (priority offer)
For whom
You used an AI coding tool, or several, and now have a codebase that almost works. It may also be code you inherited without the person who wrote it.
Problems this addresses
- The project does not install, build or start on a clean machine.
- Tests are missing, or exist and fail, or pass without testing anything that matters.
- The README describes features that the code does not have, or omits ones it does.
- Secrets, keys or personal data sit next to the source code or in the history.
- Duplicated folders, copies and branches, and nobody knows which one is canonical.
- Dependencies with known security advisories.
- Nobody can say what is safe to deploy.
What you receive
- Assessment report: inventory of what exists; what installs, builds and runs, with the exact commands and outputs; README claims versus reality; risks, sorted into "safe to change", "needs review" and "needs your approval".
- Stabilisation plan: an ordered list of fixes with an acceptance criterion for each, which you approve before implementation begins.
- Implementation of the approved items: small commits on a separate branch, never force-pushed to your main branch.
- Verification log: every test, build or check reported with its command and output. A check that was not run is listed as not run.
- Handoff: an updated README, a list of known limitations, and what remains open.
How we validate
- Nothing counts as "passing" unless it was executed; command and output go in the log.
- README statements are checked against the code and the running build, not the other way round.
- Where there is ambiguity (two candidate versions, missing source), work on that item stops and the question goes to you instead of being guessed.
- Irreversible actions (deleting branches or history, changing visibility or licences, touching secrets or production) only happen with your explicit approval.
What we do NOT promise
- We do not promise that every project can be rescued. Sometimes the assessment concludes that rewriting is cheaper; you will hear that before spending more.
- We do not promise security. We report the advisories, secrets and risks we found with the tools we ran. That is not a security audit or a certification.
- We do not promise deployment or uptime before assessing the target environment.
- We do not promise performance numbers we have not measured.
- We do not rewrite your Git history or delete anything without your written approval.
Evidence of this way of working
These are our own projects, not client rescues:
- A self-directed audit and consolidation of a full GitHub account and local workspace: inventory, duplicate detection, a secret scan over the history of 41 repositories with 0 findings, and a stop on every ambiguous project. Not paid client work.
- In comercial-template, a critical framework advisory was cleared by updating Next.js (commit 0e9e5db), and the remaining advisories are documented rather than hidden.
S02 · Custom Software & MVP Development
For whom
You need a first working version of a web application: enough to show, test with real users or build on.
What you receive
- A scoped first version: the core workflow, authentication where needed, persistence, and the screens that matter.
- Automated tests for the logic that carries risk (money, permissions, data ownership).
- A README with setup, the run command, a list of what is implemented, and the known limitations.
- A versioned repository with readable history.
- Synthetic seed data for demos, never your real data.
How we validate
- Pass/fail acceptance criteria are agreed before the build starts.
- Tests are run and their output recorded; CI is set up where the project allows.
- Limitations are written down in the delivery, not discovered later.
What we do NOT promise
- No production hosting, scaling or security guarantees unless they are scoped and assessed separately.
- No features beyond the agreed criteria inside the same scope.
Reference project
comercial-template: authentication, per-user data scoping, client and proposal workflows, server-side PDF; 24/24 unit tests passed and CI passed on the current commit. Its documented limits include no hosted deployment and an unverified Docker image.
S01 · Business Automation & AI
For whom
Anyone who handles the same intake, triage and reply work every day.
What you receive
- A mapped process: where requests arrive, how they are classified, who decides.
- A working flow: intake → automatic triage → suggested response → human approval → record.
- A clear label on every automated step stating whether it is a rule or a language model. A fixed rule is never presented as AI.
- AI providers connected only with your authorisation and account, behind an interface that can be swapped.
- Logs and an event history so you can see what happened and why.
How we validate
- Each step of the flow is tested with synthetic data.
- Nothing is sent to your customers without a person approving it.
What we do NOT promise
- No guaranteed time or cost savings. We have not measured any yet, so we do not quote any.
- No fully automatic replies to your customers.
- No AI provider costs hidden in the price: if a paid model is needed, that is your decision and your account.
Reference project
FLOWDESK is the live test demo for this offer. This rebuilt MVP uses rules-based suggestions, requires human approval, and gives each visitor a separate session for synthetic data. No AI model or real delivery is connected.
→ FLOWDESK
Not sure which one fits?
Describe the problem; the assessment decides the path.
→ Contact
Have a problem that should become software?
Describe what you have and what is not working.