LectureForge converts slide decks and PDFs into narrated, lesson-style videos — entirely on local hardware, with no external API keys required.

How it works
Upload a .pptx or .pdf file through the Next.js frontend. The FastAPI backend orchestrates the pipeline: content extraction → LLM script generation → TTS synthesis → FFmpeg video rendering. The AI engine (Ollama) runs locally inside Docker, serving either Llama 3.2:3b (fast, optimised for Apple Silicon) or Llama 3.1:8b (higher accuracy, requires 8GB+ RAM).

Architecture
LectureForge runs as a microservices ecosystem within a shared Docker network:
– Frontend (Next.js): File upload, pipeline progress tracking, and video playback.
– Backend (FastAPI): Orchestrates extraction → scripting → synthesis → rendering.
– AI Engine (Ollama): Serves the local LLM on port 11434.
– Data Volume: Persistent storage for uploads and generated video artefacts.
– Nginx: Reverse proxy configuration included.

Narrative modes
– Professor mode: Educational summaries designed for structured learning.
– Storyteller mode: Reframes content as a hero’s journey narrative.

Status: Stable containerised prototype (43 commits). CI pipeline active via GitHub Actions. Currently expanding into StoryGenerator capabilities and async Redis-backed queue processing for 50+ page documents.

Privacy: All document processing and video generation stays on your local machine when using local providers.