DEV Community

L. Cordero
L. Cordero

Posted on

Clew Quest: Intelligent Project Discovery with Algolia Agent Studio

#devchallenge #algoliachallenge #ai #agents

This is a submission for the Algolia Agent Studio Challenge: Consumer-Facing Non-Conversational Experiences

Clew Quest: Intelligent Project Discovery with Algolia Agent Studio

What I Built

Clew Quest is an open source project discovery tool that helps developers find GitHub repositories matched to their skill level and learning goals.

Instead of searching "machine learning" and getting overwhelmed by 50,000 results with no context, users answer 4 simple questions:

  • What do you want to learn?
  • What languages or skills do you care about?
  • What's your skill level? (Beginner, Intermediate, Advanced)
  • Search

They get back GitHub projects with specific explanations for why each one matches their criteria.

The tool indexes 2,800 GitHub repositories and uses semantic search to understand developer intent, not just keywords. Each result includes: project name, star count, language, difficulty level, and AI-generated reasoning explaining why it matches the user's goals.

Users can provide feedback (useful, neutral, not useful) on recommendations—data that can be used to improve future results.

Demo

GitHub: https://github.com/earlgreyhot1701D/clew-quest

Video: https://youtu.be/sP9eQe3xEA4

Audio Overview: https://notebooklm.google.com/notebook/0a3b7c43-1446-459f-94d1-868cbaa91f42?artifactId=04209d69-2163-4cfd-9f4c-8da7faeda29a

How I Used Algolia Agent Studio

Data Indexed

2,800 GitHub repositories from a public dataset, each with:

  • Project name
  • Star count & fork count
  • Primary language
  • Description
  • Difficulty level (Beginner, Intermediate, Advanced—added via Gemini enrichment)

The data pipeline: CSV → Gemini Batch API for difficulty classification → Algolia index.

Retrieval Strategy

Algolia NeuralSearch handles the retrieval layer by combining:

  • Keyword search - Matches explicit terms ("Python", "React", "machine learning")
  • Vector/semantic search - Understands developer intent ("learning ML" retrieves beginner-friendly projects, not just repos with "ML" in the name)
  • Ranking - Results ordered by relevance + community signal (star count)

The query combines user intent + selected skills + ecosystem preference to guide the search. When a user says "I want to learn Python machine learning," the search understands this as: find beginner-friendly (if selected) projects related to both Python AND machine learning, ranked by community adoption.

Prompt Engineering

The system prompt for Gemini (via Agent Studio) focuses on generating specific, actionable reasoning:

For each project result, Gemini generates a 1-2 sentence explanation that:

  1. Names the project
  2. Connects it to the user's specific intent and skill level
  3. Includes one concrete signal (star count, community size, or use case)

Example outputs:

  • "scikit-learn is ideal—top Python ML library (46K stars) with beginner tutorials everywhere"
  • "TensorFlow is the industry standard (182K stars) so you learn skills used by 90% of ML teams"
  • "Rust is perfect for systems programming—memory safety built-in, growing community"

Agent Studio Integration

Agent Studio orchestrates the entire flow in one API call:

  1. Receives user query from the React form
  2. Sends to Algolia Agent Studio with the system prompt
  3. Agent Studio coordinates:
    • Gemini analyzes the user's intent and skill level
    • Algolia NeuralSearch queries the github_projects index based on that analysis
    • Gemini generates specific reasoning for each result
  4. Returns ranked results with explanations
  5. Frontend displays project cards with reasoning, star count, language, difficulty level, and feedback buttons

The integration is seamless: one endpoint, one orchestration layer, no custom backend code needed.

Why Fast Retrieval Matters

Developer experience depends on speed and relevance working together.

Speed: Algolia's infrastructure means search results appear quickly. Slow retrieval kills discovery—users don't wait for results when searching for projects.

Relevance: Semantic search means results actually match what users are looking for. If a user searches "learning machine learning," they get beginner-friendly projects, not just repositories with "ML" mentioned in passing.

Together, these create a workflow where:

  • Users get answers immediately
  • Answers are contextually relevant to their specific goal and skill level
  • The AI reasoning is grounded in actual project data (no hallucinations)
  • Users trust the recommendations because they make sense

Without fast retrieval, the agent reasoning would be slow and frustrating. Without semantic search, users would see irrelevant results despite Gemini's reasoning being correct. Algolia enables both speed AND relevance at the same time.

GitHub: https://github.com/earlgreyhot1701D/clew-quest

Portfolio: https://beyond-the-docket-489960083310.us-west1.run.app/

LinkedIn: https://www.linkedin.com/in/la-shara-cordero-a0017a11/

Top comments (0)