What if your lecture notes could do more than just sit in a folder waiting for exam season?
That was the idea behind Recallix, a local AI-powered study assistant I built for one of my Friend as part of the DEV Community challenge.
Instead of manually summarizing lectures, extracting important concepts, and figuring out what to revise, Recallix helps turn raw notes into structured, actionable study material.
The Problem
Lecture notes can become overwhelming, especially when students have to manage multiple subjects, lengthy explanations, and revision schedules.
The challenge isn't always finding study material. Sometimes, it's making that material easier to understand, organize, and revise.
I wanted to build something that could simplify this process without adding another complicated tool to a student's workflow.
The Solution: Recallix
Recallix takes lecture notes or transcripts and uses a locally running language model to process them.
Here's what it does:
- AI-generated summaries: Converts lengthy notes into concise summaries.
- Key concept extraction: Identifies important concepts from lecture material.
- Actionable study tasks: Generates practical revision tasks based on the content.
- Contextual Q&A: Lets students ask questions about a specific lecture.
- Grounded answers: Uses the provided notes as context and refuses to answer when the information is insufficient.
- Graceful failure handling: Saves notes even when the AI model is unavailable.
- Task tracking: Helps track and complete revision tasks.
The goal is simple: spend less time organizing notes and more time learning.
Screenshots
The project includes a dashboard for managing lectures and tasks, an interface for adding lecture notes, and a contextual Q&A experience.
The screenshots are available directly in the GitHub repository, inside the docs/screenshots/ directory and README.
Tech Stack
I built Recallix using:
- Next.js, React, TypeScript and Tailwind CSS for the frontend.
- Python and FastAPI for the backend.
- SQLite for local data storage.
- Ollama for running AI models locally.
- Gemma 3:1B for summarization, concept extraction, task generation, and contextual Q&A.
The architecture keeps the application straightforward:
Frontend → FastAPI → Ollama/Gemma → SQLite-backed lecture data
One of the key decisions was to use local AI rather than depend on a paid cloud API. This keeps the application accessible without API keys and avoids sending study notes to an external AI provider.
How I Built It
I approached Recallix as a full-stack application rather than just an AI wrapper.
The backend handles lecture creation, persistence, AI processing, task management, and question answering. The frontend provides a dashboard to interact with those capabilities.
I also focused on what happens when things don't go as planned.
For example, if Ollama is offline, lecture notes are still saved. The application shows a clear warning instead of breaking completely.
I tested the core workflows, including lecture processing, grounded question answering, unsupported questions, and AI failure handling.
The backend test suite passed 27 tests, and the frontend linting and production build completed successfully.
What I Learned
Building Recallix helped me explore several practical aspects of AI-powered application development:
- Integrating a locally running language model into a full-stack application.
- Designing prompts for structured AI output.
- Building contextual question answering around user-provided information.
- Handling unsupported questions rather than blindly generating answers.
- Designing graceful failure behavior when an AI dependency is unavailable.
- Connecting frontend, backend, database, and AI components into one usable workflow.
It also reinforced an important lesson: a useful AI application isn't just about getting a model to generate text. Reliability, usability, and knowing when the model doesn't have enough information matter just as much.
What's Next?
Recallix is currently a local application. There are several directions in which it could grow:
- PDF and audio lecture ingestion.
- Editing and deleting lectures.
- Exporting summaries and revision tasks.
- Searching across lecture content.
- Streaming responses in the Q&A interface.
These are possible future improvements, not features currently implemented.
Try It Yourself
Recallix runs locally and requires Ollama with the Gemma 3:1B model.
The repository contains the installation instructions, setup commands, project structure, and screenshots.
GitHub: https://github.com/namandeeptripathi/Recallix
There is currently no public hosted demo because the application relies on local model inference.
Built with curiosity, local AI, and the goal of making studying a little more organized.
Top comments (0)