🧠 Testing the RAG Pipeline
After building the initial RAG workflow, we needed to verify an important question:
Is the AI actually receiving the right information from the patient's documents?
We tested the complete flow:
Medical Document
↓
Amazon S3
↓
Amazon Textract
↓
Extracted Text
↓
Chunking / Processing
↓
Retrieval
↓
RAG
↓
AI Response
The goal was not simply to generate an answer, but to make sure the response was based on the relevant information retrieved from the uploaded documents.
We tested different questions and checked whether the expected document information was being passed through the pipeline correctly.
🎨 Most of Day 3: Improving the UI
And honestly...
The UI took most of our time today. 😅
A feature can work perfectly from the backend, but if the user experience is confusing, the application still doesn't feel complete.
We spent a significant amount of time refining the CareRelay frontend.
Things we worked on:
- Improving the overall layout
- Making the document section easier to navigate
- Improving the AI chat interface
- Making responses easier to read
- Improving spacing and component organization
- Handling different UI states
- Connecting frontend interactions with the backend APIs
- Fixing small visual and functional issues
Some changes looked like they would take only a few minutes.
They didn't. 😂
One small UI change often affected another component, which meant testing, debugging, and adjusting the implementation again.
🔗 Connecting the UI With the AI Pipeline
Another important part of the work was making sure the frontend and backend worked together smoothly.
The intended flow is:
User
↓
CareRelay UI
↓
API Gateway
↓
Backend / Lambda
↓
Patient Documents
↓
RAG Pipeline
↓
Relevant Context
↓
AI Response
↓
CareRelay UI
This helped us identify issues that aren't always visible when testing the backend independently.
🧪 Build → Test → Fix → Repeat
Day 3 reminded us that building an AI application isn't only about the AI model.
We had to continuously move between:
Backend → RAG → API → Frontend → Testing → Debugging → UI
A working feature is only one part of the process.
Making that feature understandable and usable is another challenge altogether.
☁️ AWS Technologies Behind CareRelay
Our current CareRelay architecture continues to build on AWS services:
- Amazon S3 — Medical document storage
- Event Bridge - To perform textraction
- Amazon Textract — OCR and document text extraction
- AWS Lambda — Serverless backend processing
- Amazon API Gateway — API layer
- Amazon DynamoDB — Application/document metadata
- RAG + AI layer — Document-grounded question answering
👥 Building With the Team
Although I am participating in the Bharat Builds Tour as a solo participant, I have teammates who support me with debugging, implementation issues, and technical discussions.
Having people to discuss problems with makes a big difference when you're stuck on something that seems simple but refuses to work. 😅
📈 Day 1 → Day 2 → Day 3
Our progress is starting to take shape:
Day 1
📄 Document upload and storage
Day 2
🧠 OCR + RAG + AI + Initial UI
Day 3
🧪 RAG testing + major UI improvements
Each day is turning CareRelay from an idea into a more complete working application.
🎯 Day 3 Takeaway
Today's biggest lesson was simple:
Building the feature is only the beginning. Making it reliable and easy to use takes just as much work.
Most of our time today went into the UI, but that work helped us understand how the entire CareRelay system behaves from a user's perspective.
We're still building, testing, breaking things, fixing them, and learning along the way.
Day 3 complete. 🚀
Build → Test → Improve → Repeat.
Loading article