Workwisecase study / Workwise
An AI tender-analysis pipeline for EPC contractors
A private, per-tenant document pipeline on AWS that reads tender documents end to end and turns them into a brief estimators can review: scope, eligibility, BOQ structure, risk clauses and deadlines.
- Client
- Workwise
- Industry
- Construction tech / SaaS
- Service
- Generative AI
About the client
Workwise is a connected project-management and AI platform for EPC and industrial contractors in India. It covers a project from tender and estimation through procurement and execution to closeout.
letsworkwise.com (opens in a new tab)The challenge
Every bid starts with the tender, and EPC tenders are long: hundreds of pages of scope, eligibility conditions, bills of quantities and contract clauses, often scanned. Workwise needed to turn any tender PDF into a short brief its customers' estimators could review, fast enough to change how bids get prepared. Because tenders are commercially confidential, each customer's documents had to stay isolated and in India, never be used to train shared models, and leave an audit trail of every AI-assisted action.
What we built
- Tender uploads land in Amazon S3 under per-tenant prefixes, each encrypted with that tenant's own KMS key, in an AWS Region in India
- An AWS Step Functions workflow takes every document through text extraction, chunking, analysis and assembly, with retries and a dead-letter path for documents that fail
- Amazon Textract asynchronous analysis pulls text and tables, including BOQ tables, out of large and scanned PDFs
- Lambda functions split each tender into sections and run Amazon Bedrock prompts that extract scope, eligibility, BOQ structure, high-risk clauses and submission dates
- Results are stored per tenant in DynamoDB and served to the Workwise app through API Gateway, with Cognito tokens scoping every request to a single tenant
- Every AI-assisted action is logged with the user, timestamp and source document, with CloudTrail and CloudWatch dashboards across the pipeline
- All infrastructure is defined in Terraform, so new environments are reproducible
Architecture
AWS services used
| AWS service | Role in the solution |
|---|---|
| Amazon S3 + AWS KMS | Stores uploaded tenders under per-tenant prefixes, each encrypted with the tenant's own key. |
| AWS Step Functions | Orchestrates each document through extraction, analysis and assembly, with retries and failure handling. |
| Amazon Textract | Extracts text and tables, including BOQ tables, from large and scanned tender PDFs. |
| AWS Lambda | Splits tenders into sections and runs the analysis prompts, with no servers to manage. |
| Amazon Bedrock | Generates the tender synopsis and extracts BOQ structure, risk clauses and deadlines, without training on customer data. |
| Amazon DynamoDB | Holds each tender's brief and processing state, partitioned by tenant. |
| Amazon API Gateway + Cognito | Serves results to the Workwise app, with every request scoped to one tenant. |
| Amazon CloudWatch + AWS CloudTrail | Pipeline monitoring and alerts, plus an API-level audit trail. |
Outcomes
- A 600-page tender read end to end in about four minutes (Workwise's own published figure), giving estimators a brief to review instead of a document to read
- Each customer's tenders isolated, encrypted with their own key and stored in an AWS Region in India
- Customer documents never used to train shared models
- An audit trail for every AI-assisted action: who ran it, when, and on which document
More case studies
next step
Ready to scale smarter?
Book a free 30-minute call with one of our engineers. Bring your architecture, your bill or your AI idea, and leave with a plan for what to fix first.