Recall solves a very real student problem: we study across too many places, but we do not retain a usable memory of what we actually learned.

A student might watch a YouTube lecture, open Google Classroom, read BYJU’S, check The Helper for SRM notes and PYQs, open a PDF from Downloads, and maybe browse Coursera or NPTEL. After a few hours, all of that becomes scattered across tabs, history, files, and random notes. Revision then becomes stressful because the student does not know:

• what they actually covered
• which topic is weak
• what to revise first
• which source is trustworthy enough to study from again

Existing tools mostly do one of two things: either they are generic note/summarization tools that need manual input, or they are cloud-heavy assistants that do not understand the student’s actual study trail.

Recall solves this by acting like a local-first AI study memory layer. It captures real study activity from trusted educational sources, filters out distractions and low-signal content, turns study material into short notes, flashcards, revision prompts, and syllabus-aware guidance, and then shows the student a clear next step instead of a pile of raw information.

For SRM students, this is especially useful because learning is spread across classroom portals, PYQs, The Helper, local notes, and online lectures. Recall turns that scattered learning into something structured, explainable, and actually reusable for exams, projects, and placements.

If you want, use these too for the other Devfolio sections:

What it does
Recall is a browser-based, local-first AI study memory extension that captures real educational activity and converts it into structured study support. It can understand trusted study sources like YouTube lectures, Google Classroom, BYJU’S, Coursera, NPTEL, SWAYAM, SoloLearn, The Helper, and local files like PDFs, PPTs, DOCX, audio, and video lectures. It generates readable notes, visual flashcards, quiz prompts, revision guidance, syllabus mapping, placement insights, and an explainable audit trail of why a source was accepted or rejected.

Inspiration
As students, we realized the real problem is not lack of content, but lack of continuity. We keep learning from many different platforms, but we lose track of what we studied, what matters, and what to revise. We wanted to build something that feels like a personal academic memory system instead of just another chatbot or note app.

How we built it
We built Recall as a browser extension with a popup, dashboard, document studio, and AI mentor. The system uses local-first AI logic for educational source detection, topic extraction, note generation, revision planning, and flashcard generation. We also built custom local scoring systems like SourceGuard for study-source trust and CardRank for flashcard quality. For deeper semantic analysis and offline speech transcription, we integrated on-device models through Transformers.js and Whisper. We also added an optional Python AI backend for broader assistant behavior and future extensibility.

Accomplishments that we're proud of
We are proud that Recall is not just a demo chatbot. It is a full product flow with trusted source capture, readable notes, guided study sessions, flashcards, auditability, SRM-focused syllabus support, local AI models, and an optional Python AI backend. We are especially proud of SourceGuard and CardRank because they make the product feel unique and proprietary instead of generic.

What we learned
We learned that “AI for students” is only useful when it is grounded in the real study workflow. We also learned that explainability matters a lot: students and judges both trust the product more when it can show why it captured something, why it rejected something, and why it recommends a certain revision path.

What’s next for Recall
Next, we want to improve mixed-format note quality, strengthen lecture understanding further, expand SRM department-specific packs, improve the Python AI teaching mode, and make the product even more personalized through adaptive study modes and stronger offline AI support.