video1.0<iframe src="https://www.loom.com/embed/7da1d3f3c2b0418f8e8f3adc125c7be4" frameborder="0" width="1280" height="960" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>9601280Loomhttps://www.loom.com9601280https://cdn.loom.com/sessions/thumbnails/7da1d3f3c2b0418f8e8f3adc125c7be4-048e94f23c35c0de.gif498.226Hi, I'm Gaurav Deshmukh . This is my submission for the Eywa SDE Intern assignment — a Telegram bot that summarizes YouTube videos and answers questions, powered entirely by LLaMA 3 running locally via Ollama, so no paid APIs and no data leaving your machine. The stack is simple: Python handles the Telegram layer, youtube-transcript-api fetches captions, and all AI processing goes through Ollama on localhost with per-user session management and 2-hour transcript caching. Let me show it live — I'll paste a YouTube link, the bot fetches the transcript and returns a structured summary with key points, timestamps, and a core takeaway. I can then ask questions about the video and it answers strictly from the transcript — no hallucination — and if a topic isn't covered, it says so explicitly. I can also say 'Summarize in Hindi' and it instantly regenerates in Hindi, with support for Tamil, Kannada, Telugu, Marathi, and Bengali as well. There are bonus commands too — /deepdive for full analysis, /actionpoints for practical takeaways, and /language to switch anytime. Full code and setup instructions are on GitHub. Thanks for watching.