reimon01/Tutor-Virtual-ITSON-Sistema-Multi-Agente-RAG

Telegram bot serving as a virtual academic tutor for ITSON students: answers about curriculum, academic calendar, study techniques and emotional support using a vectorized knowledge base in Supabase. The repo implements a controlled A/B experiment comparing dynamic retrieval (RAG) vs. static knowledge injected in the system prompt; both response types are persisted for comparative analysis. Key workflows: - INGESTARAG: watches Google Drive, extracts PDF text and vectorizes with Cohere embeddings into Supabase. - RAG: Telegram bot with retrieval from vector store, conversational memory, user profiling, and concurrency control. - LLM: same bot using hardcoded system prompt as baseline. Stack: n8n orchestration, Supabase (Postgres + pgvector), OpenAI gpt-4.1-mini and Groq llama-4-scout-17b, Cohere embeddings, Telegram frontend. Notable features: per-user locks in Supabase with stale-lock cleanup, profiling sub-agent updating user profile every 4 messages, multi-LLM strategy and multiple API keys for rate-limit scaling, smart splitter for long replies. Workflows are sanitized; replace credential placeholders and create four Supabase tables (documents, messages, user_context, active_executions). See docs/ARQUITECTURA.md for schema and deploy steps.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.