SHL Assessment Recommender

An AI-powered conversational API that recommends SHL assessments based on job role requirements through multi-turn dialogue.

Live API

Endpoint	URL
Health	GET https://shl-assessment-recommender-agent.onrender.com/health
Chat	POST https://shl-assessment-recommender-agent.onrender.com/chat
Docs	https://shl-assessment-recommender-agent.onrender.com/docs

Quick Start

Health Check

curl https://your-app.onrender.com/health

API Reference

POST /chat

Request:

Response:

Test Type Codes:

Architecture

Evaluation Results

Proof of terminal output

Per-Trace Performance

Local Development

Prerequisites

• Python 3.14.0

• GROQ API key (free at console.groq.com)

• Create requirements.txt file, all packages listed in this file check it in my git repo.

Setup

Clone

• git clone https://github.com/YOUR_USERNAME/shl-assessment-recommender.git
• cd shl-assessment-recommender

Virtual environment

• python -m venv venv
• source venv/bin/activate # Windows: venv\Scripts\activate

Install

• pip install -r requirements.txt

Environment

• echo "GROQ_API_KEY=your_key_here" > .env

To remove ChromaDB or any file

• rmdir /s /q chroma_db_v10

Run

• uvicorn app.main:app --reload --port 8000

Run Evaluation

• python shl/Scripts/test_traces.py

Project Structure

Tech Stack

Step 6: Render Deployment (15 minutes)

Step 7: After Deploy - Update README URLs

Replace YOUR_APP_URL with actual Render URL

Then push again

• git add README.md
• git commit -m "Update README with live API URL"
• git push

Screenshots

Packages Required

FastAPI uvicorn Server Logs

Agent given recommendations of SHL Assessments with Groq LLM reply.

Hosted in Render.com platform checked both API endpoints GET /health, POST /chat

Using curl commands tested API endpoint of POST /chat

Out of SHL Assessemts means this SHL Assessment Recommender Agent don't know Off topics which is irrelevent to the SHL Assessments.

getting 10 recommendation from the SHL Catalogue according to the query.

Using https://shl-assessment-recommender-agent.onrender.com/docs it provides Auto Swagger UI to test both API endpoints GET /health & POST /chat

POST /chat

Achievement

10 Public Conversation traces I trained the AI AGENT LLM model improvement of Mean Recall@10: 0.8757 means 88% and Behavior Pass Rate: 10/10 means 100.0%, Validation Schema: 10/10, Per-Trace Recalls runs for each trace has turns and get 10 recommendations through SHL-Catalogue Dataset in JSON format.