🤝 Top Collaborators
.png&w=640&q=75)
.png&w=640&q=75)
.png&w=640&q=75)
.png&w=640&q=75)
🤓 Latest Submissions
Aachari
Aachari is a local, source-grounded AI educational video generation platform designed to automate the process of turning a learning topic or source material into a structured, evidence-backed educational video. Instead of relying on a simple prompt-to-video workflow, Aachari uses a multi-stage generation pipeline where research, content generation, asset creation, spatial layout, rendering, validation, and release are handled as separate governed stages. The system maintains PostgreSQL as the durable source of truth, allowing every generation run and its intermediate checkpoints to be tracked and reproduced. How Aachari works A generation request first creates a frozen input snapshot and a database outbox record. The pipeline can either use a user-provided source or retrieve authoritative web pages through its research route. These sources are persisted as source documents so that generated educational content can remain grounded in evidence. The generation workflow is organized into 16 governed stages, beginning with preflight and progressing through research, content and asset generation, spatial layout, rendering, validation, and finally a release record. Each stage has its own handler and defined input dependencies, while the pipeline executor manages stage leases, heartbeats, checkpoints, and retry classification. Aachari also separates AI-generated decisions from deterministic operations. For example, final overlay coordinates are not directly trusted from an LLM. Instead, deterministic SVG diagrams provide measured anchors, and a spatial solver calculates and verifies the final placement. This helps make the generated video layout reproducible and reduces unpredictable visual positioning. Technical architecture The project is implemented as a monorepo containing applications, infrastructure, packages, scripts, tests, and documentation. Its backend pipeline uses PostgreSQL for durable state, BullMQ/Valkey for job dispatching, MinIO for private artifact storage, provider adapters for AI services, and Remotion for headless video rendering. The model/provider system is also configurable. Model routing is maintained through a provider registry, allowing individual stages to be routed to different provider/model combinations. The resolved model route is stored with the stage checkpoint, helping preserve reproducibility even if environment configuration changes later. Main objective The core objective of Aachari is to create an end-to-end, reliable and reproducible educational-video production system, rather than simply generating a video from a single AI prompt.
27 Sep 2026
