After endlessly fidgeting and not finishing my project, I finally decided to start a project.
Manga-torial
Manga-torial is a SaaS application that transforms dense college course material into engaging, personalized, manga-style comic strips. It leverages a fully serverless cloud architecture, intelligent Retrieval-Augmented Generation, and a metered subscription model.
💻 Tech Stack
| Domain | Technology | AWS services |
|---|---|---|
| Frontend | Next.js | AWS Amplify Hosting |
| Backend API | Python (google-genai) SDK, (psycopg2) | AWS Lambda & API Gateway |
| Database | PostgreSQL (pgvector) | Amazon RDS (serverless v2) |
| AI/ML | Gemini API | N/A |
| Authentication | AWS Amplify SDK | AWS Cognito |
| Billing/FinOps | Stripe API (Webhooks) | AWS Lambda & API Gateway |
| Storage | Not Applicable | Amazon S3 |
✨ Key Features
1. Serverless RAG Pipeline
- Document Ingestion: upload material -> AWS S3 -> trigger Lambda function
- Vectorization: Lambda chunks text and generates vector embeddings -> store in PostgreSQL using pgvector
- Contextual Storytelling: user asks concept -> RAG model retrieves most relevant vector-embedded content.
2. Personalized AI Generation
- Structured Output: Lambda creates prompts with RAG context -> enforce Gemini to output structured story script
- Multimodal Output: Gemini generates comic -> saves to S3
- Dynamic UI: parse S3 url and create responsive manga-style viewer
3. Monetization & Security
- Subscription Gating: Integrate Stripe using AWS Lambda web hook -> update user database upon payment
- Tiered Rate Limiting: API Gateway enforces blanket rate limit on all requests
- Internal Gating Lambda checks user's subscription status in PostgreSQL -> returns 429 Too many Requests when quota exceeded
- Secure Access: All API endpoints are secured by Amazon Cognito Authorizer
TL&DR
I'm starting Serverless project to turn study material to content using AI
💡
I will begin by designing the database schema. See you in the next blog!