Tender management software development
AI that reads the tender, and shows you the page.
Tender management software takes in tender packages, pulls out the line items, quantities, and bond terms, and tracks each bid to a decision. Acasa Labs builds it with AI extraction that refuses to guess. SigmaCPM, our construction platform, cites the tender page for every value it extracts and drops any value it cannot find there.
The problem
An estimator's week goes on copying, not pricing.
Tender packages run to hundreds of pages, often scanned, and the line items an estimator needs are spread across schedules, specifications, and annexures. Copying them into a spreadsheet takes days, and the bond terms hide in the conditions of contract. Generic AI tools will summarise a tender happily and invent a figure along the way. In a bid, one invented number can cost more than the tool ever saved.
What we build
- 01
Intake for scanned and digital tenders
OCR with retry handling for scanned pages, then extraction of line items and terms across the whole package.
- 02
Every value cited
Each extracted value carries a quote and the page it came from, so the estimator checks the source instead of searching the PDF.
- 03
Bond terms checked twice
A second model run re-reads the bond terms, and any disagreement is flagged for a person.
- 04
A bid pipeline
Every tender tracked from intake to submission and result, with the extracted items feeding vendor quote requests and the estimate.
From production
How SigmaCPM reads a tender.
The quote must be on the page
SigmaCPMFor every value, the model must return a quote, and the system checks that the quote exists on the cited page. If it does not, the value is dropped, not guessed.
An adversarial second pass
SigmaCPMBond terms are re-checked by a second model run. When the runs disagree, the term lands in a disputed state for a person to resolve.
Read the right pages, not all of them
SigmaCPMCandidate pages are retrieved before extraction, so the model reads the pages likely to hold the answer instead of the whole package.
Cost on every call
SigmaCPMTokens and cost are recorded on every LLM call, so the price of reading a tender is known as it happens, not estimated after the invoice arrives.
Honest trade-offs
When AI tender extraction is not worth it
If you bid a few tenders a year from short, clean documents, an estimator with a spreadsheet is fine. Extraction pays off when you bid often, when packages are long or scanned, or when the same items go on to vendor quotes, bonds, and billing. Even then, the AI does the reading and your estimator makes every decision. We build it that way on purpose.
What we reach for
- OpenRouter LLMs
- OCR
- Java PDFBox
- Django REST Framework
- Postgres job queue
- Next.js
Proof
Shipped, not pitched.
01Our product
SigmaCPM
Construction project management where AI reads the tender and shows its work.
One project record for civil contractors, from tender to closeout: bids, contracts, RFIs, vendor quotes, and drawings. The AI pulls line items and bond terms out of tender PDFs and cites the page for every value.
Web · API · Chrome extension
Questions
Tender management software questions we get asked.
How accurate is AI tender extraction?
We do not quote an accuracy figure, because it depends on your documents. What we can describe is how SigmaCPM handles errors: every value must be quoted from its cited page or it is dropped, and bond terms get a second check.
Can it read scanned tender documents?
Yes. Scanned pages go through OCR, with retry handling, before extraction.
Which details can it extract?
SigmaCPM extracts line items and bond terms. A custom build can target the fields your team needs, such as deadlines, eligibility criteria, or penalty clauses, under the same quote-and-cite rule.
Which AI models does it use?
SigmaCPM calls models through OpenRouter, so the model can change without rewriting the pipeline, and it records token use and cost on every call.
Can it work with government e-procurement portals?
Tender documents downloaded from a portal go straight in. Pulling them from a portal automatically depends on what that portal allows, so we check it case by case.
Can our estimators correct the extracted values?
Yes, and they should. In SigmaCPM, disputed and uncited values already go to a person. A custom build can let the estimator edit any value with its source page alongside.
Tender management software development
Send us a tender package.
A real one you have bid on. We will show you what the extraction returns, page by page, and scope the system around your team.