I built HerTutorAI because My friend wouldn't stop asking me for Youtube recommendations.

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

I built HerTutorAI, a multimodal study assistant that takes our 5th (personalized) semester syllabus topic or lecture diagram and instantly generates AI breakdowns, study resources and curated YouTube video recommendations for those topics in one place

I built this for my friend who keeps asking me for youtube video recommendations and notes or references for our semester examinations topics. now she can just enter the topic (from our 5th semester syllabus) and get explanation and recommended video on that topic, it also provides references , from where to study the topic.

Problem it solved

1. Eliminates "Search Fatigue": Instead of searching YouTube blindly,my friend enters a syllabus prompt or uploads a lecture diagram. HerTutorAI parses the core concept and automatically matches it with curated, relevant educational video resources.
2. Multimodal Diagram Analysis: If a lecture slide or textbook diagram is confusing, they can drop the image directly into the chatbot to get a step-by-step breakdown alongside matching lecture video recommendations.
3. 100% Offline & Fast: Built with local LLMs (Ollama) and local compiled Tailwind CSS, it operates completely privately without subscription paywalls, API costs, or bandwidth throttling.

Demo

View demo on github

Code

Check out the full open-source codebase on GitHub:
👉

zainab-ansari-dev / HerTutorAI

HerTutorAI 🎓

Built for a friend who asked for an easy way to organize learning resources, lecture breakdowns, and curated YouTube videos in one place.

HerTutorAI is an offline-capable, multimodal academic study assistant created for Hacktoberfest DEV Challenge 'Build for a Friend'. It combines local LLM inference, pre-computed UI card rendering, and automated YouTube video curation into a fast, privacy-first learning hub.

Why I built this

I built this for my friend who keeps asking me for youtube video recommendations and notes or references for our semester examinations topics. now she can just enter the topic (from our 5th semester syllabus) and get explanation and recommended video on that topic, it also provides references , from where to study the topic.

Problem it solved:

1. Eliminates "Search Fatigue": Instead of searching YouTube blindly,my friend enters a syllabus prompt or uploads a lecture diagram. HerTutorAI parses the core concept and…

View on GitHub

How I Built It

Open-Source AI Technologies Used

• Open-Weight Models:

  • Llama 3.2 (1B): Used as our core reasoning model for generating structured explanations, breaking down complex academic concepts, and pre-computing active-recall flashcard data.
  • LLaVA (Vision-Language Model): Used for multimodal visual understanding to analyze uploaded lecture slides, notes, and textbook diagrams step-by-step.

• Local Inference Engine:

  • Ollama: Serves as our local model runner, allowing HerTutorAI to run 100% offline with zero API keys, no monthly SaaS fees, and complete user privacy.

• Backend Integration & Framework:

  • FastAPI (Python): Communicates asynchronously with Ollama via standard local HTTP endpoints, enabling real-time response streaming and non-blocking background tasks.

---

How HerTutorAI is Built Around Open-Source AI

HerTutorAI was architected from the ground up to prioritize offline privacy, zero latency costs, and local compute:

1. Local Model Orchestration: Rather than relying on cloud LLM APIs (like OpenAI or Anthropic), LockIn AI routes all text requests to local llama3.2 instances and visual requests to local llava instances managed by Ollama.
2. Streaming & Structured Extraction: As Ollama streams tokens locally, our FastAPI backend parses the stream to extract structured JSON data for topic tags without delaying the user experience.
3. Data Privacy & Accessibility: By keeping the entire AI pipeline local and open-source, we can study without an internet connection, keeping their notes and study materials completely on their own device.

Why Open Innovation Matters for HerTutorAI

Open innovation is the backbone of HerTutorAI. If this project relied on closed, proprietary cloud APIs (like OpenAI or Anthropic), it would completely destroy its core promise : unlimited, zero-cost, private offline learning.

Here is what open-weight models (Llama 3.2 & LLaVA via Ollama) made possible that closed APIs couldn't:

1. Zero Financial Barrier for Students: Closed APIs charge per-token fees that scale with usage. generating hundreds of flashcards, explanations, and diagram breakdowns daily, API costs quickly become unsustainable. Open-source models cost $0 to run locally.
2. 100% Offline Capability & Zero Search Fatigue: Closed APIs require an active, stable internet connection and strict rate limits. HerTutorAI works seamlessly in dorms, libraries, or low-connectivity zones, keeping all study workflows local.
3. Complete Privacy & Data Ownership: Students often paste copyrighted lecture notes, personal class slides, or practice exams into AI tools. With open-source AI running locally, no student data or study material ever touches an external third-party server.
4. Full Architectural Control: Closed APIs are black boxes—their parameters, system prompts, and endpoint behaviors can change overnight. Open-weight models gave us complete freedom to customize prompt structures, orchestrate local multi-model routing (Llama 3.2 for text + LLaVA for vision), and stream pre-computed flashcards without relying on cloud availability.