Schema inference
Works out what data your problem needs before you go looking for it.
The general-purpose AI optimiser.
Describe your problem in plain English and upload your data in any reasonable format. AI agents turn it into a formal model, then a traditional solver finds the answer. The AI sets the problem up, but it never makes up the solution.
Your description"Assign 100 students to 4 forms. Keep friends together, keep rivals apart, and balance ability and gender across every form."
Objective
Maximise friendships keptMinimise ability spread
Hard rules
Max 30 per formRivals separated
Preferences
Balanced genderEven ability
Decisions
Student → Form
Built with
OR-ToolsPydantic AIFastAPIPostgreSQLClaude
Try a tiny version.
Eight pupils, two forms of four. Pick your rules, press Optimise, and see what comes back. This small demo runs in your browser, so it can check every possible split. Real problems are far too big for that, which is where the solver and agents come in.
The problem
Hard rule
Preferences
The result
Optimisation works, but it's usually out of reach.
Every organisation has to share out limited people, time and resources under competing rules. Doing that well has meant consultants, fragile spreadsheets and weeks of modelling.
The old way
- Weeks of expert modelling before you see an answer
- Data cleaned by hand into one exact shape
- Solutions nobody can explain or safely challenge
- Back to square one for every small change
With Screet
- Write the problem down and get a working model in minutes
- Upload data in any reasonable format and agents map it for you
- Every solution comes with a plain-English explanation
- Adjust, pin and re-run, and the solver carries on from there
From problem to solution in four steps.
The agents handle the translation and the paperwork. You stay in charge of every decision that matters.
1
Describe the problem
Explain it the way you would to a colleague. The agent picks out your objectives, the hard rules that can never be broken, the preferences to meet where possible, and what is actually being decided. It plays all of this back for you to confirm before anything runs.
2
Supply the data
The agent works out what data the problem needs, then reads what you have: CSV, JSON, spreadsheets or documents. Mismatched columns, inconsistent formatting and missing fields are mapped and checked automatically, and anything unclear is put to you first.
3
Optimise
Your problem is compiled into a formal model and tested on generated sample data, so surprises show up early. Then it runs on your real data with live progress. If the rules can't all be met, Screet tells you straight away and names the ones in conflict.
4
Refine
Read a plain-language summary of the result, edit it by hand, pin the parts you want to keep, or ask for targeted improvements. Re-run whenever you like and the solver picks up from the current state.
Agents suggest. The solver decides.
Language models are good at understanding messy requests and spotting patterns, but they can get things wrong and can't prove an answer is the best one. So in Screet the AI never writes your solution. It sets up the model and proposes moves, and a traditional solver such as OR-Tools builds, checks and scores every answer.
- No invented answers
Every solution is produced by the solver, and your hard rules are checked against your real data, not taken on an AI's word.
- Agents help when it stalls
When progress levels off, an agent suggests swaps for the solver to try, to get it out of a rut it can't see its way out of.
- Every move is checked
The solver tests each suggestion and keeps only those that respect the hard rules and improve the result.
Why "Screet"? It stands for Systematic Cooperative Resolver with Exact Engine Technology: agents and solver working together, with the exact engine having the final say.
How the two take turns
- SolveThe solver builds a strong first solution.
- ReviewAn agent reads it and spots rules that are still unmet.
- SuggestIt proposes targeted swaps, and the solver checks each one.
- ResumeThe solver restarts from the improved state and carries on.
Repeats until results stop improving.
What you get along the way.
Automatic validation
Catches missing values, wrong types and clashing records before the solver runs.
Sample-data test runs
Solves a generated dataset first, so problems surface before a real run.
Pinning and warm starts
Lock in the decisions you trust and re-run without starting from scratch.
Plain-English results
Each solution comes with a readable summary of what was met and what gave.
Changes by conversation
Ask for a different priority in words and the model updates to match.
Export and reuse
Export to CSV, JSON or a report, and save a setup to refine next cycle.
Full audit trail
Every input, run, adjustment and pinned decision is recorded for review.
If you can describe the rules, it can use them.
Screet suits any problem where limited things have to be shared out under rules and preferences.
Workforce and rostering
Shift patterns, skills and availability, shared out fairly across large teams.
Education and timetabling
Class lists, rooms and courses, balancing ability, friendships and capacity.
Logistics and routing
Deliveries, vehicles and loads planned within time and cost limits.
Resource allocation
Budgets, equipment or people split sensibly across competing priorities.
Built by a small team, and already tested in schools.
Some of the most valuable decisions an organisation makes, like who works when, which child goes in which class, or where a limited resource ends up, are still made by hand. The techniques to solve them properly have been around for decades. What's been missing is the expertise to apply them.
Screet puts rigorous optimisation behind a simple, conversational interface, so the person who understands the problem can solve it themselves. We've run trials in educational institutes with success, and we're now opening early access to teams with real problems to try it on.
Founder, ScreetMost organisations are already solving an optimisation problem. They just don't have the words or the tools for it. We want everyone to have what used to sit behind a consultant's desk.
Bring us your hardest problem.
We're letting in a small group of early-access teams. Leave your email and we'll be in touch when a place opens up.
Please enter a valid email address.
Thanks, you're on the list. We'll be in touch when early access opens.
By signing up you agree to us emailing you about early access, and nothing else. See our privacy notice.
← Back to Screet
Privacy notice
Last updated 11 October 2026
This notice explains what happens to your details when you use screet.co.uk. We keep it short because we collect very little.
Who we are
Screet is a startup division of Traject Ltd (company number 10264028). Traject Ltd is the data controller for the personal data described here, and "we" and "us" in this notice mean Traject Ltd.
What we collect
- Your email address, if you enter it in the early-access form.
- Standard server logs. Our web server records your IP address, the page requested and the time of each visit, which we use for security and to keep the site running.
This site doesn't use cookies, analytics or advertising trackers.
Why we use it
We use your email only to contact you about Screet early access. We don't send marketing for other things and we never sell or rent your details. Our legal basis is your consent, which you give by submitting the form. You can withdraw it at any time.
Who sees it
Your email is stored on a server we run ourselves and is only accessible to the Screet team. The site loads its fonts from Google Fonts, so your browser sends your IP address to Google when the page loads. We don't share your email with Google.
How long we keep it
We keep your email until you ask us to delete it, or until the early-access programme ends, whichever comes first. Server logs are kept for a short period and then deleted.
Your rights
Under UK data protection law you can ask to see the data we hold about you, have it corrected or deleted, or object to how we use it. Email [email protected] and we'll respond within a month. If you're unhappy with how we've handled your data, you can complain to the Information Commissioner's Office.