Foundations
Architecture first
Separate AI service, queue workers and a vector store designed up front, so the product could scale without a rewrite.
TendIT: AI that reads a tender and drafts the answers from your own knowledge base.
Responding to RFPs is slow, repetitive and high stakes. TendIT reads the documents, finds the questions and drafts grounded answers, then scores how good they are.
- RAG
- grounded answers
- Monthly
- iteration
- Scored
- answer quality
The problem
Every tender starts from a blank page, even when the answers already exist.
Bid teams rewrite the same answers for every RFP, and quality depends on who is available that week. The founder wanted a system that reads the tender, pulls the right material from the company's own documents and drafts a first answer worth editing.
What we built
- Upload tender documents; questions extracted automatically
- Knowledge base of the client's past answers and documents
- Grounded answer drafting, with sources
- Answer quality scoring pipeline
- Real-time updates, billing and team workspaces
Technology
What it runs on.
Vue 3
Laravel 12 API
Python FastAPI AI service
Qdrant vector DB
Pusher
AWS S3
Stripe
How AI is used
Retrieval-augmented generation. Tender documents and the client's knowledge base are embedded into a vector database. OpenAI GPT models extract the questions and draft answers grounded in that material, and an evaluation pipeline scores answer quality so the team can see what needs a human.
How we worked
Product and AI evaluation led by Growth Pilot, built in monthly blocks.
Blocks
Extract, draft, score
Question extraction first, then grounded drafting, then quality scoring. Each reviewed live on real tenders.
Now
Monthly iteration
A monthly release rhythm, with the evaluation pipeline guiding what to improve next.
Where it is now
Live, iterating monthly.
In use on real tenders, with new releases every month.
Talk to us about an AI productBuilding an AI product and need it grounded in reality?
Start with a Blueprint: 3 to 5 days, and a plain go or no-go call at the end.