Back to Projects

ResilienceBOT

Agentic-RAG emotional coaching chatbot: 3 orchestrated n8n pipelines, a 28-emotion detection API and a full RAG pipeline with safety rules against hallucination.

Demo

Overview

ResilienceBOT is an Agentic-RAG emotional coaching chatbot developed as a PFA (final-year project) at ENIT, in co-authorship. It orchestrates 3 dedicated n8n pipelines — chat, ingestion and lifecycle — with multi-turn memory and session persistence, so conversations feel continuous and context-aware.

On the AI side, a 28-emotion detection API built on distilroberta-base normalises predictions into 7 coaching tones and is served via FastAPI. A complete RAG pipeline combines Supabase pgvector as a vector store, Google Drive as a knowledge source and Google Gemini for generation.

Appreciative-coaching safety rules are embedded directly in the system prompt, with mandatory vector anchoring to reduce hallucination — ensuring ethical, personality-tailored responses for every user.

n8nSupabase pgvectorGeminiFastAPIPython

Key Features

  • 3 orchestrated n8n pipelines (chat, ingestion, lifecycle)
  • Multi-turn memory with session persistence
  • 28-emotion detection API (distilroberta-base → 7 tones) served with FastAPI
  • Full RAG pipeline: Supabase pgvector + Google Drive + Google Gemini
  • Appreciative-coaching safety rules embedded in the system prompt
  • Mandatory vector anchoring to reduce hallucination

Have a project in mind?

I'm open to internships, collaborations and exciting engineering challenges.

Get in Touch