Bibintell: Focus Buddy Chrome Extension

Bibin caught me slacking from physics. Pop-up window shows focus stats
Bibin

Inspiration

As engineering students, when we are overwhelmed by a huge influx of homework or when we are pulling all-nighters like the one at GenAI Genesis, we are often burnt out and lose focus from our task at hand. Have you ever found yourself switching between tabs, opening mails and messages, evening going on scroll breaks. JUST ANYTHING BUT FOCUSED!

We tried website blockers, but their indifferent barricade makes turning them off frictionless. We tried systems that cultivate affinity with the user through mechanisms like pets and gardens, but their lack of enforcement makes bypassing the tracker easy.

This is why we built Bibintell -- your judgemental productive buddy who develops attachment with the user.

What it does

Bibintell reimagines digital productivity by replacing rigid, punitive site blockers with Bibin, an AI-powered beaver who acts as your browser-based study companion. Instead of simply cutting off access to the web, Bibintell uses real-time semantic analysis to understand the context of your browsing. When you start a session, you tell Bibin your goal—whether it's organic chemistry or 18th-century history— and he follows along as you research. If the AI detects that you’ve drifted from your syllabus into a rabbit hole of unrelated content, Bibin pops up with a personalized, conversational nudge to steer you back. It’s a supportive, psychological approach to focus that culminates in a detailed stats dashboard where users can track their streaks, focus percentages, and top distractions.

How we built it

We engineered Bibintell using a FastAPI backend and a Manifest V3 Chrome Extension to ensure a persistent, lightweight user experience. To make Bibin’s interventions feel "human" rather than robotic, we implemented a two-stage AI pipeline: first, Sentence Embeddings and cosine similarity provide a high-speed relevance filter; if sustained drift is detected, a more nuanced LLM reasoning agent (gpt-oss-120b) is triggered via HuggingFace Inference Endpoints to verify the distraction. Moreover we fine-tuned Gemma 270M as the conversation engine to generate a context-aware nudge that matches the personality of Bibin. We chose Supabase for our database to handle session logging and real-time stats, while the extension’s service worker manages the complex lifecycle of background timers and cross-tab messaging. This architecture allowed us to combine the heavy-lifting of RAG and semantic analysis with the seamless responsiveness required for an in-browser assistant.

Challenges we ran into

Git merge conflicts - 3 hours of process being erased Google chrome extension not running on some of our computers Background removal of animated videos Adding supabase

Accomplishments that we're proud of

We designed a original character, Bibin with detailed description. Water colour style art enables seemless reminder and quiet presence of Bibin.

Working with locally hosted LLM (Ollama)

Working with Google frontend when we all have no clue what even is a middleware.

What we learned

AI pipelines to generate usable animations - GPT-Image for keyframes and WAN2.2 for frame connection. You need to understand front-end to vibe code front-end. Test before git pull --rebase