Case study · Developer tools (SaaS): A RAG chatbot platform that lets developers ship AI assistants in minutes
- Client
- A UK developer-tools startup
- Industry
- Developer tools (SaaS)
- Country
- United Kingdom
- Year
- 2025
- Timeline
- 16 weeks to launch
- Cloud
- AWS
System architecture
How the pieces connect
- AWS
- Live flow
01Sources
- PDF & DOCX uploads
- Website crawl
- Notion & Confluence
02Ingest
- Amazon S3AWS
- Amazon SQSAWS
- AWS LambdaAWS
03Index
- Bedrock embeddingsAWS
- OpenSearch ServerlessAWS
04Answer
- ECS on FargateAWS
- Amazon BedrockAWS
- Amazon DynamoDBAWS
05Deliver
- API GatewayAWS
- CloudFront widgetAWS
- Amazon CognitoAWS
(01)Results
What changed.
10
AWS services in production
1,200+
Chatbots created in the first six months
< 2s
Median time to a cited answer
(02)The challenge
The problem.
The client wanted to give developers a faster way to add AI assistants to their products. Their early prototype answered questions from uploaded documents, but it ran on a single server, mixed every customer's data in one index and slowed down badly once more than a handful of teams used it at the same time.
To launch commercially they needed strict separation between customers' data, answers that cite their sources, predictable model costs, and an architecture that could absorb spikes when a customer's chatbot went viral, all without growing an in-house infrastructure team.
(03)The solution
What we built.
We rebuilt the product as a serverless, multi-tenant platform on AWS. Developers create a workspace, connect their sources (PDFs, websites, Notion and Confluence exports), and get a chatbot they can test in a playground, then embed with a script tag or call through a REST API.
Uploaded content flows through an asynchronous ingestion pipeline that cleans, chunks and embeds each document, then stores the vectors in a per-tenant index. At question time the platform retrieves the most relevant passages, applies the workspace's instructions and guardrails, and streams an answer with links back to the source documents.
Multi-source ingestion
PDF, DOCX, web crawls and knowledge-base exports, with scheduled re-sync so answers stay current.
Tenant isolation
Separate indexes, encryption keys and API credentials for every workspace.
Embeddable widget & API
A lightweight chat widget served from the edge, plus a streaming REST API for custom front ends.
Analytics & feedback
Question logs, thumbs-up/down feedback and an 'unanswered questions' report that shows which docs to improve.
(04)Architecture
10 cloud services, one system.
AWS
Amazon Bedrock
Foundation models for answer generation and embeddings
Amazon OpenSearch Serverless
Vector search with a separate index per tenant
Amazon S3
Encrypted storage for uploaded source documents
Amazon SQS
Queue that buffers ingestion jobs during bulk uploads
AWS Lambda
Parsing, chunking and embedding of documents
Amazon ECS on Fargate
Chat and admin APIs with streaming responses
Amazon API Gateway
Public API with keys, rate limits and usage plans
Amazon DynamoDB
Workspaces, chat history and usage metering
Amazon Cognito
Developer sign-up, login and team roles
Amazon CloudFront
Global delivery of the dashboard and chat widget
Stack
- TypeScript
- Next.js
- Node.js
- Python
- LangChain
- Terraform
- GitHub Actions
(05)The outcome
After launch.
The platform launched to paying customers after 16 weeks. Because every component scales on demand, the client absorbed several large traffic spikes without manual intervention, and infrastructure costs track usage instead of sitting idle.
Developers can go from uploading documents to a working, embedded chatbot in a single session, which became the client's main selling point. The same pipeline now supports new source types and models without changes to the core platform.
“9Falcons took a fragile prototype and turned it into a platform we could confidently sell. The architecture has scaled with us without a single rewrite.”
Services in this project
More case studies
Want results like these?
Tell us what you want to build or automate. We'll reply with honest advice and a clear next step.