Resolva – AI Customer Support Agent by ABDULLAH SHAFIQUEResolva – AI Customer Support Agent by ABDULLAH SHAFIQUE

Resolva – AI Customer Support Agent

ABDULLAH SHAFIQUE

Completed work

AI Developer

Web Developer

React

Resend

Vite

Computer Software

Resolva — AI Customer Support Agent

Confidence-based AI support with automatic human escalation. Built on a 100% free stack for portfolio/demo use.

Stack

Layer Tool LLM Groq (LLaMA 3.1 70B) — 14,400 req/day free Embeddings sentence-transformers (local, CPU) Vector DB ChromaDB (local persistent) Agent LangGraph 0.3.x Backend FastAPI + SQLAlchemy + SQLite Email Resend.com (3,000/month free) Frontend React + Vite

Quick Start

1. Backend

Backend runs at http://localhost:8000 API docs at http://localhost:8000/docs

2. Frontend

Frontend runs at http://localhost:5173

3. Ingest your knowledge base

In the app, go to Upload KB tab:

Set Company ID (e.g. my-company)

Set API Key (matches UPLOAD_API_KEY in .env)

Upload a .pdf or .txt file

Or via curl:

4. Chat

In the app, go to Chat tab:

Set Company ID to match what you uploaded

Ask questions — confident answers go through, low-confidence ones escalate

Environment Variables

Variable Required Default Description GROQ_API_KEY ✅ — Get free at groq.com RESEND_API_KEY ❌ — For escalation emails HUMAN_AGENT_EMAIL ❌ — Who gets escalation emails UPLOAD_API_KEY ✅ changeme Protects the upload endpoint CONFIDENCE_THRESHOLD ❌ 0.65 0.0–1.0, higher = more escalations DATABASE_URL ❌ sqlite:///./resolva.db SQLAlchemy URL CHROMA_DB_PATH ❌ ./chroma_db Vector store path MAX_UPLOAD_SIZE_MB ❌ 10 Max upload file size

API Endpoints

Method Path Description POST /chat Send a message, get AI response POST /upload Upload knowledge base document GET /tickets List escalated tickets PATCH /tickets/{id}/resolve Mark ticket resolved GET /analytics Get stats (total, escalated, avg confidence) GET /health Health check

Deployment Note

Warning: Railway/Render free tier has ephemeral storage. ChromaDB and SQLite data resets on every restart. For persistence without cost, use Chroma Cloud free tier and swap PersistentClient for HttpClient in backend/rag/embedder.py.

How Escalation Works

Confidence is calculated from:

Semantic similarity between answer and retrieved docs (50%)

Semantic relevance between answer and original query (30%)

Number of docs retrieved (20%)

Like this project

Completed work

Posted Sep 28, 2026

Developed an AI customer support agent system using a AI tech stack.

Likes

Views