Inspiration
For 16 years I ran a small cleaning business in South Africa and regularly saw how difficult it could be for a small contractor to navigate the tender process.
The problem wasn't simply finding a tender. Tender documents can be long, complicated, and difficult to interpret. Important requirements, scoring criteria, deadlines, forms, supporting documents, and pricing information can be spread across many pages. Missing one required item can potentially make all the other work irrelevant.
Large companies may have people and resources dedicated to procurement. A small contractor often has to figure it all out alone.
That experience inspired Tender Clarity: an AI-assisted tool designed to help smaller contractors turn a complicated tender document into something they can actually understand and act on.
The goal isn't to promise someone a winning bid. The goal is to give them clarity, structure, and a better-prepared starting point.
What it does
Tender Clarity takes one tender PDF and turns it into a practical preparation workspace.
It extracts and organizes important information and separates findings into three categories:
- STATED — information explicitly found in the tender.
- CONCLUSION — an interpretation or conclusion derived from information in the tender.
- UNCLEAR — information that could not be established confidently and needs attention.
Where possible, findings include the relevant page and source excerpt so the contractor can check where the information came from.
The tool also provides:
- A tender overview
- An evidence-backed preparation checklist
- Progress tracking for checklist items
- Next actions
- A ZAR cost worksheet covering labour, materials, transport, equipment, overheads, and other expenses
- A comparison between estimated costs and an entered contract value
- A downloadable PDF report containing the current results
The prototype deliberately keeps the contractor in control. It does not submit a tender, replace a procurement authority, or guarantee a successful bid.
How we built it
I am not an experienced software developer, so this project was built using an AI-assisted, plan-first workflow.
Instead of immediately asking AI to write an application, I worked through the problem in stages:
- Define the problem and narrow the scope.
- Create a product requirements document.
- Create a technical specification.
- Build and test the application in small verified slices.
- Use automated tests and Git checkpoints throughout the build.
The prototype uses a deliberately simple architecture:
- Python
- Streamlit for the web interface
- pypdf for page-by-page PDF text extraction
- Gemini as the AI analysis provider
- ReportLab for downloadable PDF reports
- Git for checkpoints and version control
The application keeps the current analysis, checklist, and costs in the active browser session rather than introducing a database or complicated infrastructure.
A provider adapter also keeps the AI layer separate from the rest of the application so the architecture can evolve later without rebuilding the entire product.
Challenges we ran into
One of the biggest challenges was designing the product so that AI didn't simply produce a nice-looking summary that a contractor couldn't verify.
That led to the evidence model separating what the tender actually states from conclusions and unresolved information.
We also needed to make sure source excerpts really corresponded to the extracted tender text. Automated tests were created to check these references, including findings supported by information across multiple pages.
Another challenge was building within strict resource constraints. I am working with an older, limited laptop and have no budget for paid development infrastructure. We therefore kept the architecture small and designed the prototype around free tools and Free Tier usage.
During development, the live Gemini service also returned HTTP 503 (UNAVAILABLE). Rather than hiding that failure or automatically switching to a paid service, the application reports the problem clearly and does not enable a paid fallback. The rest of the prototype can continue to be tested using controlled data.
Accomplishments that we're proud of
The biggest accomplishment is that an idea that started as a personal frustration has become a working proof of concept.
The prototype now has:
- Source-grounded tender analysis
- STATED / CONCLUSION / UNCLEAR evidence categories
- Source and excerpt verification
- An interactive preparation checklist
- Checklist progress tracking
- A ZAR cost viability worksheet
- Cost and margin calculations
- Handling for below-cost contract values
- A complete downloadable PDF report
- Automated tests covering the major functionality
At the current checkpoint, the automated test suite has passed all 22 tests.
More importantly, the project demonstrates a workflow that could potentially be useful beyond the hackathon: helping a person with limited procurement resources understand a complex opportunity before deciding how to respond.
What we learned
We learned that building with AI is much more effective when the AI is used as a structured development partner rather than simply being asked to generate an entire application.
The plan-first workflow forced us to make decisions about the user, the problem, the scope, the evidence model, and the technical architecture before building.
We also learned that AI-generated conclusions need to be treated carefully. A system can provide a useful interpretation while still being wrong, which is why Tender Clarity makes the underlying tender evidence visible instead of presenting every AI conclusion as fact.
Finally, we learned that a small, tested application is more useful at this stage than a technically complicated system with many unfinished features.
What's next for Tender Clarity
The next major capability we want to explore is tender intelligence.
One particularly valuable direction is researching historical information about similar or previous tenders, including publicly available contract awards and pricing information where that information can be reliably found.
That could eventually give small contractors additional context when estimating costs and evaluating an opportunity.
Other future possibilities include broader document support, stronger web research, more sophisticated tender comparison, and eventually making the tool available to organisations that genuinely want to help small businesses participate more effectively in procurement.
For this hackathon, however, the focus remains deliberately small:
Take a complicated tender, make it clearer, show the evidence, help the contractor prepare, and help them understand the numbers.
Log in or sign up for Devpost to join the conversation.