Answering tenders from a lifetime of bid documents

A PropTech business working with Cloud Combinator on AWS. The client is anonymised at their request.

The proof of concept had three focus areas, each tied to the project's success criteria.

Source documents are stored in Amazon S3 and ingested by an Amazon Bedrock Knowledge Base, which uses the Titan Embed G1 model to convert them into vector embeddings. Those embeddings are stored and indexed in Amazon Aurora PostgreSQL acting as the vector database, enabling fast semantic search across the whole library.

The engagement delivered a working retrieval-augmented bid-writing pipeline in the client's AWS account: document ingestion and embedding through Amazon Bedrock Knowledge Base, a vector store in Aurora PostgreSQL, and a Bedrock Agent served through API Gateway and Lambda, deployed as infrastructure as code. It demonstrated AWS as a suitable environment for the client to continue developing the service.

• Knowledge made retrievable — Years of documents in S3 were embedded and indexed so the right content can be found by meaning, turning a static library into a searchable knowledge base.
• Question answering on tap — A Bedrock Agent connected to the knowledge base interprets a natural language question and returns a contextually relevant answer, with the goal of covering the bulk of a tender questionnaire.
• Focused, consistent scope — Concentrating first on Housing Association and the platform tenders kept responses consistent and avoided the added complexity of Build to Rent bids.
• Conversational continuity — Chat history and session tracking in DynamoDB let the assistant hold context across a series of questions.
• Handover and documentation — The team received full technical and user documentation and a handover, along with guidance on switching Bedrock models to tune performance.

With a validated pipeline and a clear knowledge base in place, the client has an AWS foundation on which to widen coverage across tender types, refine answer quality and put faster, more consistent bid responses within reach of the whole team.