🚀 Make Presentation Simple

Transform PDFs into structured presentation decks using a fully serverless, AI-powered pipeline.

🌐 Live App: https://vakt-artifactbucket-vxgd6cj5bqrc.s3.us-east-1.amazonaws.com/index_.html

---

✨ Overview

This project is a serverless AWS application that converts PDFs into presentation-ready slide decks using:

• 📄 PDF parsing
• 🖼️ Figure extraction
• 🤖 Multimodal + LLM processing
• ☁️ Fully managed AWS services

It leverages AWS SAM, Step Functions, and Lambda to create a scalable, event-driven pipeline.

---

🧠 High-Level Design (HLD)

Architecture Summary

• Frontend (SPA) → Upload & interaction
• API Gateway → Entry point
• Lambda Functions → Business logic
• Step Functions → Workflow orchestration
• S3 → File storage
• DynamoDB → Job tracking
• Hugging Face Models → AI generation

🔷 HLD Diagram

Paste this directly into a Markdown file that supports Mermaid:

flowchart LR
    U[User] --> FE[Frontend SPA]

    FE -->|POST /upload| API[API Gateway]
    FE -->|GET /status/:id| API
    FE -->|POST /preprocess/:id| API
    FE -->|POST /process/:id| API

    API --> CU[Lambda: create_upload]
    API --> ST[Lambda: status]
    API --> SP[Lambda: start_preprocess]
    API --> SPR[Lambda: start_processing]

    CU --> DDB[(DynamoDB JobTable)]
    CU --> S3[(S3 ArtifactBucket)]
    ST --> DDB

    SP --> SM1[Step Functions: Preprocess]
    SPR --> SM2[Step Functions: Process]

    SM1 --> ET[Lambda: extract_text]
    SM1 --> RP[Lambda: render_previews]

    SM2 --> CF[Lambda: crop_figures]
    SM2 --> CM[Lambda: call_model]
    SM2 --> WJ[Lambda: write_final_json]
    SM2 --> CL[Lambda: cleanup_data]

    ET --> S3
    RP --> S3
    CF --> S3
    CM --> S3
    WJ --> S3
    CL --> S3

    ET --> DDB
    RP --> DDB
    CF --> DDB
    CM --> DDB
    WJ --> DDB
    CL --> DDB

Loading

---

---

⚙️ Low-Level Design (LLD)

Core Workflows

1️⃣ Upload Flow

Paste this directly into a Markdown file that supports Mermaid:

sequenceDiagram
    participant User
    participant FE as Frontend
    participant API as API Gateway
    participant CU as create_upload Lambda
    participant DDB as DynamoDB
    participant S3 as S3

    User->>FE: Select PDF
    FE->>API: POST /upload
    API->>CU: create upload request
    CU->>DDB: Create job record
    CU->>S3: Generate presigned URL
    CU-->>FE: upload_id + upload_url
    FE->>S3: Upload PDF

Loading

2️⃣ Preprocessing Flow

sequenceDiagram
    participant FE as Frontend
    participant API as API Gateway
    participant SP as start_preprocess
    participant SM as Step Functions
    participant ET as extract_text
    participant RP as render_previews
    participant DDB as DynamoDB
    participant S3 as S3

    FE->>API: POST /preprocess/{id}
    API->>SP: Start preprocessing
    SP->>SM: Trigger workflow
    SM->>ET: Extract text
    ET->>S3: Save raw.txt
    ET->>DDB: Update status
    SM->>RP: Generate previews
    RP->>S3: Save images
    RP->>DDB: Update previews

Loading

3️⃣ Processing Flow

sequenceDiagram
    participant FE as Frontend
    participant API as API Gateway
    participant SPR as start_processing
    participant SM as Step Functions
    participant CF as crop_figures
    participant CM as call_model
    participant WJ as write_final_json
    participant CL as cleanup_data
    participant DDB as DynamoDB
    participant S3 as S3

    FE->>API: POST /process/{id}
    API->>SPR: Start processing
    SPR->>SM: Trigger workflow
    SM->>CF: Crop figures
    CF->>S3: Save crops
    CF->>DDB: Update state
    SM->>CM: Generate slides (AI)
    CM->>S3: Save final.json
    CM->>DDB: Update state
    SM->>WJ: Write viewer
    WJ->>S3: Save index.html
    WJ->>DDB: Save viewer URL
    SM->>CL: Cleanup
    CL->>S3: Delete temp files
    CL->>DDB: Mark complete

Loading

---

🧩 Key Components

Component	Role
create_upload	Initializes job + presigned URL
extract_text	Extracts PDF text
render_previews	Generates page images
crop_figures	Crops user-selected regions
call_model	AI slide generation
write_final_json	Publishes final output
cleanup_data	Removes temporary files

---

☁️ AWS SAM Deployment

Prerequisites

• AWS CLI configured
• AWS SAM CLI installed
• Python 3.12

🔧 Build

sam build

🚀 Deploy

sam deploy --guided

During deployment, provide:

• Hugging Face API token
• Stack name
• Region

🔁 Subsequent Deployments

sam deploy

---

📦 Storage Structure (S3)

uploads/<upload_id>/
  ├── original.pdf
  ├── raw.txt
  ├── previews/
  ├── figures/
  ├── final.json
  └── index.html

---

🔄 State Management (DynamoDB)

Each job tracks:

• state (waiting, processing, done)
• progress
• previews
• viewer_url

---

⚠️ Notes

• The frontend currently triggers /process/{upload_id} immediately after upload (even for preprocessing).
• The backend separates preprocessing and processing workflows.

---

💡 Why This Project Matters

• Fully serverless → no infrastructure management
• Scalable pipelines via Step Functions
• Multimodal AI integration
• Real-world document understanding use case

---

📜 License

MIT License