CogniOps

Inspiration

Cloud computing is one of the highest-demand skills in tech, yet most learners — especially from Tier-2 and Tier-3 cities in India — are stuck watching passive tutorials with no real guidance. Developers also waste hours switching between documentation, architecture tools, cost calculators, and debuggers. CogniOps was built to fix that — an AI mentor powered by Amazon Bedrock that takes a learner from "what is AWS Lambda?" all the way to generating production-ready Terraform infrastructure.

What it does

CogniOps is a dual-mode AI cloud platform — persona locked at registration.

Student Mode (CogniBot Student — Powered by Claude on Bedrock)

• AI Chat mentor that gives real-time answers, generates 8-week personalized roadmaps, and creates visual AWS architecture diagrams on request
• Quiz Generator with difficulty levels — scores tracked and shown on dashboard
• Progress Dashboard with XP points, day streak, modules done, and avg score
• Voice input support (mic button) for hands-free learning
• Multilingual support — EN/TA/HI language toggle

Developer Mode (CogniBot Dev — Powered by Claude on Bedrock)

• AI Chat that recommends full AWS architecture stacks for any app idea
• Architecture Generator — describe your app, get Overview + visual Diagram + Terraform + Cost breakdown in 4 tabs
• Backend Designer — paste your code, get recommended AWS services, API endpoints, database suggestions, and implementation notes
• Terraform Generator — describe infrastructure, get production-ready HCL code with copy button
• Socratic Debug Assistant — guides you to the fix with diagnostic questions, likely causes, and step-by-step fix instructions

How we built it

Frontend: Flutter Web — hosted on Amazon S3

Backend: AWS Lambda (Python) + Amazon API Gateway

AI Engine: Amazon Bedrock — Claude Sonnet 4.6

Auth: Amazon Cognito

Database: Amazon DynamoDB

Monitoring: Amazon CloudWatch + AWS IAM

All requests from Flutter hit API Gateway → Lambda → Bedrock. User progress and chat history are stored in DynamoDB. Cognito handles role-locked authentication.

Challenges we ran into

• Dual persona in one codebase: Two completely different toolsets in one Flutter app required careful Provider-based state management and role-aware Lambda functions.
• Prompt engineering per feature: Every screen needed its own AI personality — the debugger guides without giving answers, the Terraform generator outputs real production code, not pseudocode.
• Serverless cold starts: Initial Lambda latency was high. Mitigated through provisioned concurrency, keeping average AI response time at 2–4 seconds.
• Flutter Web + Cognito: Managing auth tokens across Flutter Web required careful CORS handling at API Gateway.

Accomplishments that we're proud of

• Fully deployed live app on AWS with real Amazon Bedrock responses — not a mockup
• End-to-end serverless architecture that scales automatically
• 2–4 second average AI response time for architecture generation and concept explanations
• Scalable to 1000+ concurrent users at ~$120/month

What we learned

• Prompt design is product design — the same Bedrock model gives drastically different quality based on system prompt structure
• Flutter Web on S3 is a serious, fast, cheap deployment path
• Serverless-first architecture forces good stateless design decisions
• Role-locking at registration eliminates an entire class of conditional UI logic

What's next for CogniOps

• Multilingual voice interaction using Amazon Transcribe (Hindi + regional languages)
• Visual AWS architecture diagram generation as SVGs
• One-click Terraform + CloudFormation export from Developer Mode
• Dedicated Android/iOS mobile app

Built With

• amazonapigateway
• amazonbedrock
• amazondynamodb
• amazoniam
• amazonlambda
• amazons3
• awscognito
• dart
• flutter