EdgeSlate: Offline AI Presentation Pipeline

EdgeSlate is an end-to-end, ethical and transparent AI presentation generation pipeline designed for offline rural educational environments. The system ingests structured textbook PDFs, routes contextual payloads through a local edge-computing LLM (phi3:mini), and compiles responsive HTML5 and PPTX format presentation layers embedded with kinetic typography animations.

Once initialized, the entire pipeline operates 100% locally and offline, shielding sensitive educational materials from external networks.

---

Architecture

NCERT PDF
    │
    ▼
┌─────────────────────────────────────────┐
│  Layer 1: Material Ingestion            │
│  PyMuPDF span extraction → Font         │
│  hierarchy detection → Level 1 LDA →   │
│  SQLite (edgeslate.db)                  │
└────────────────────┬────────────────────┘
                     │ topic chunks
                     ▼
┌─────────────────────────────────────────┐
│  Layer 2: PCCG Framework                │
│  Level 2 LDA → Prompt construction →   │
│  phi3:mini @ localhost:11434 →          │
│  4-criterion validation gate →          │
│  Temperature-scaling retry →            │
│  Deterministic fallback                 │
└────────────────────┬────────────────────┘
                     │ validated JSON
                     ▼
┌─────────────────────────────────────────┐
│  Layer 3: Dual Output Rendering         │
│  Branch A: PPTX (python-pptx +         │
│            matplotlib + pyttsx3)        │
│  Branch B: HTML (Base64 + inline SVG)  │
└─────────────────────────────────────────┘

---

The PCCG Framework Novelty

Standard LLM pipelines generate output and use it directly. PCCG generates output and validates it against four pedagogical criteria before accepting it:

Criterion	What It Checks
Structural validity	Output is valid JSON with title, bullets, narration fields
Cognitive load	Bullet count between 3–4; no bullet exceeds 25 words
Placeholder detection	No generic titles, no bracket-enclosed instructions
Hallucination signals	No phrases indicating the model deferred content generation

If a slide fails any criterion, the system retries with a linearly scaled temperature:

T_i = 0.3 + (i × 0.15),  i = 0, 1, 2, 3, 4, 5

This gives 6 attempts at temperatures 0.30 → 0.45 → 0.60 → 0.75 → 0.90 → 1.05.

If all 6 attempts fail, a deterministic sentence-extraction fallback activates: sentences are split from the source NCERT text using regex, filtered by length, truncated to 12 words, and used directly as bullets. The fallback guarantees that a slide is always produced and is always factually grounded in the source text.

This architecture decouples reliability from model capability, a 3.8B parameter quantized model produces 100% reliable structured output through constraint enforcement rather than scale.

Deployment Directory Structure

When evaluating the extracted deployment package, the directory should contain the following core assets:

EdgeSlate_windows/
├── setup.bat           # Environment provisioning & AI engine initialization script
├── ui.exe              # Main compiled application binary (GUI)
├── edgeslate.db        # Pre-configured relational SQLite knowledge repository
├── assets/             # Core system graphics, UI layout, and structural templates
├── output/             # Target output directory for generated standalone HTML assets
└── _internal/          # Encapsulated native runtime dependencies

Installation & Deployment Workflow

Phase 1: Environment Initialization (Internet Required)

To ensure the local machine has the necessary AI infrastructure to process documents offline, an initial provisioning step is required. This step only needs to be performed once.

1. Extract the provided ZIP package to a local directory (e.g., your Desktop). Do not run the application directly from the compressed archive.
2. Navigate into the extracted EdgeSlate_windows/ folder.
3. Execute the setup.bat script by double-clicking it.
4. A terminal window will initialize and perform the following automated actions:
   • Detect or download the local Ollama inference engine.
   • Pull the highly compressed 3.8B parameter language model (phi3:mini).
5. Please wait for the process to reach 100%. Depending on network bandwidth, this may take a few minutes.
6. Upon successful completion, the script will automatically terminate and launch the core application interface.

Phase 2: Standard Operation (100% Offline)

For all subsequent everyday usage, network connectivity is completely disabled and unnecessary.

1. Open the EdgeSlate for teachers folder.
2. Double-click the EdgeSlate.exe binary to launch the control dashboard.
3. Use the interface to upload your PDF materials, select chapters, and generate lesson plans.

Usage

1. Click Browse and select an NCERT Science PDF
2. Wait for extraction to complete (one-time per book, ~4 minutes)
3. Select a class level using the segment buttons
4. Select a chapter from the dropdown
5. Check one or more topics from the list
6. Set the subtopics slider (2–5 slides per topic)
7. Choose Export as PPT or Export as Web Presentation
8. Click Generate Presentation
9. Click Open Generated Output when done

Output Retrieval

Every time the generation engine compiles a new lesson, it dynamically writes a standalone, interactive HTML file or a PPTX file.

• Location: All generated presentations are automatically saved inside the output/ folder located in the root directory.
• Execution: These files do not require specialized software. Simply double-click any generated .html file to present the interactive lesson natively in any modern web browser (Chrome, Edge, Firefox) or the .pptx file to present in MSPowerpoint or LibreOffice.

Scope

Parameter	Value
Textbook corpus	NCERT EVS Classes 3–5 and Science Classes 6–7
Language	English editions only
Hardware target	Intel i3/i5, 8 GB RAM, no GPU
Network requirement	Fully offline after installation

---

Limitations

• The source coverage metric checks noun-phrase overlap, not whether causal relationships are correctly preserved
• No field evaluation with rural teachers has been conducted; all results are from controlled technical testing
• Audio narration uses Windows SAPI5, which is not calibrated for Indian English pronunciation norms
• Scanned or image-only PDFs (no selectable text layer) are not supported

Summer Internship Project — NIT Tiruchirappalli, May–June 2026 Sriranjani Karthikeyan — B.Tech CSE, VIT University, Vellore