AI-Powered Educational Web Assistant (RAG-Based)

Team name: TEAM IFI

Use of AI tools :

• Which tool(s) did you use?

  Gemini / DeepMind AI Assistant (for coding and ideation), Groq Whisper API (for Voice-to-Text), and Llama 3 (for the core RAG generation).

• For what purpose(s)?

  We used AI for structuring our initial ideas, drafting architectural concepts, generating boilerplate code, debugging complex integrations (like KaTeX for math formulas and MediaRecorder for asynchronous audio), and refining our final pitches and infographics.

• How did these tools support your work?

  They acted as an "always-available" technical co-founder. This allowed us to iterate incredibly fast, moving from a basic concept to a fully functional, mobile-first Progressive Web App with advanced features like Voice STT and Socratic tutoring in a matter of days.

External feedback & contributions :

• Jean Dersoir (CHANCE Team) – Jury/Expert – Feedback & Discussion: Raised critical questions regarding academic data security, governance, and maintaining performance in low-connectivity environments.

• Evaluation Committee / Peers – Comments & Feedback: Highlighted concerns regarding the risk of students using AI to cheat (cognitive bypass), the amplification of biases in single-source documents, and the technical challenge of poorly scanned PDFs.

Initial contribution: AI-Powered Educational Web Assistant (RAG-Based)

Final contribution: https://drive.google.com/file/d/1HK_94vf23n9byg_uRDKJ2O7uIDY9a9Eb/view
Application link : https://huggingface.co/spaces/hamba-ho/Assistant-Web-Educatif

Reflection on the process (important)

Explain how your contribution evolved during Phase 2:
In Phase 1, our project was a standard RAG (Retrieval-Augmented Generation) chatbot designed to answer questions based on a PDF. During Phase 2, the project evolved into a highly specialized, ethical, and infrastructure-aware educational platform tailored for resource-constrained environments (specifically Africa). We shifted our focus from simply "providing answers" to "facilitating learning" and ensuring digital inclusion.

• What feedback (comments, discussions, expert inputs, keynotes) influenced your work?
  The most impactful feedback came from concerns about pedagogical ethics and technical reality. The committee pointed out that giving direct answers encourages cheating, and that relying on a single PDF can amplify biases. Furthermore, Jean Dersoir challenged us on data governance and the realistic performance of AI on low-end smartphones with unstable 2G/3G networks.

• What changes did you make as a result?
  Based on this feedback, we developed four major technical updates:

  1. The Socratic Mode: We gave control back to the teachers by adding a toggle. When active, the AI refuses to give direct answers and instead asks guiding questions to make the student think.

  2. Comparative Reading Prompting: The LLM was re-engineered to explicitly detect and highlight contradictions between different course documents to build the student's critical thinking.

  3. Intelligent OCR Alert: To handle the reality of bad document scans, the backend now calculates character density during upload and instantly warns the teacher if the PDF is unreadable by the AI.

  4. Asynchronous Multilingual Voice: To respect low bandwidths, we avoided live audio streaming and built a "WhatsApp-style" asynchronous voice note system using Whisper, which perfectly handles local dialects (like Lingala or Wolof) without requiring a massive internet connection.

• How did this process strengthen your final proposal?
  This iterative process transformed our project from a generic technological demo into a mature, context-aware product. By directly translating ethical and infrastructural concerns into concrete code features (Socratic Mode, PWA architecture, asynchronous voice), our final proposal is far more resilient, scalable, and pedagogically sound. It proves that we didn't just build an AI; we adapted AI to the specific realities of African universities.

Video pitch :

https://drive.google.com/file/d/1OUNIPLltRPu0cJPEVKb2prdMv8bjiybn/view