04Tender review
Jean Claude
Reads the whole tender, so the estimator misses nothing.
- My role
- Product, discovery, spec and research
- Platform
- Mac and Windows
- Status
- Client project
- Year
- 2026
Illustration with made-up data
A tool for estimators at a Danish construction company. It reads the tender documents and points out gaps, contradictions and items in the wrong trade, with a reference to where it says so.
How it started
It started with a conversation with the CEO.
He was frustrated by how long they spent on tenders. And by how high the risk was.
A tender can be hundreds of files: specifications, drawing lists, bills of quantities and reservations.
It takes days to read and get an overview of. Pricing the bid correctly is hard, and the financial risk is high.
I suggested a different route: let an LLM read it all in a controlled, regulated way, and see how much knowledge we could pull out. Jean Claude grew from there.
Today Jean Claude reads all of it.
Jean Claude finds gaps, holes and contradictions in the material, so the estimator can set the right price.
The problem
The estimator has to get through all of it before the bid is due, often on a tight deadline. If a requirement hides in an appendix nobody reached, it's the contractor who pays.
The insight
To set the right price, we let the estimator do what he is good at, and let an LLM do the heavy chewing.
The solution
- 01Tender analysis: where documents contradict each other, what is missing and what could cause trouble in the price.
- 02Bills of quantities: build them for subcontractors straight from the tender material. It saves time and makes sure everything the tender asks for is in the lists.
- 03Subcontractor bids: what is included, what is left out, and who has the best price for the job?
- 04Tables and files are read locally. AI is only used where something needs understanding and comparing.
The principle
Everything has to be read. Nothing may be cut or skipped. The value lies in the confidence that all of it is included.
Where it stands
Built close to the estimators' daily work. The first test was a tender with 161 files and more than 800,000 characters. Today it has been used on 20 tenders.
- files in the test tender
- 161
- hours saved per tender
- c. 70
- tenders answered
- 20
Stack
- Tauri 2
- Rust
- React
- TypeScript
- Claude API
- Supabase