Skip to main content

​
Welcome to Quest

Quest is a Retrieval-Augmented Generation (RAG) engine designed to assist with solving and explaining Data Structures and Algorithms (DSA) problems. It leverages a combination of retrieval-based methods and generative models to provide accurate and context-aware solutions to coding problems, explanations of concepts, and metadata about LeetCode-style questions.

Installation

Get started by installing Quest and its dependencies

Quickstart

Run your first query in minutes

API Reference

Explore the RAG Engine API and components

Configuration

Configure retriever and generation parameters

​
What is Quest?

Quest combines the power of semantic search with large language models to provide intelligent assistance for coding problems. Instead of relying solely on web searches or generic AI responses, Quest retrieves relevant solutions from a curated dataset of 1800+ LeetCode problems and generates context-aware explanations.

​
What problems does Quest solve?

Quest uses a hash map for instant retrieval of exact problem matches, returning solutions in milliseconds rather than seconds.
When exact matches aren’t found, Quest uses FAISS and HNSW indexing to find semantically similar problems and solutions.
Quest doesn’t just return code—it generates detailed explanations incorporating conversation history and retrieved context.
All processing happens locally using Ollama, ensuring your code and queries remain private.

​
Key features

Exact matching retrieval

Quickly retrieves solutions for exact matches of problem titles from a curated dataset using optimized hash maps.

Fast similarity search

Uses FAISS and HNSW indexing for efficient similarity search in high-dimensional embedding spaces.

Dual model architecture

Switches between qwen2.5-coder:1.5b for general queries and deepseek-r1:7b for complex reasoning tasks.

Conversation memory

Retains past interactions (configurable up to N queries) for contextually accurate responses across multiple queries.

Dynamic prompt switching

Implements model-specific prompt templates to prevent infinite reasoning loops and optimize response quality.

Metadata-rich responses

Provides problem difficulty, related topics, companies, and edge cases for retrieved solutions.

Optimized performance

Reduced query retrieval time from 1 minute to 15 seconds and reasoning time from 20 minutes to under 4 minutes.

Confidence-based fallback

Switches to model’s own inference for queries with confidence below 0.6 for better accuracy on unknown problems.

​
How it works

Quest uses a multi-stage retrieval and generation pipeline:
  1. Query normalization - Normalizes user queries to check for exact matches
  2. Exact matching - Uses hash map lookup for instant exact problem matches
  3. Semantic retrieval - Encodes query using all-MiniLM-L6-v2 and searches HNSW index
  4. Confidence filtering - Filters results based on similarity scores (default threshold: 0.6)
  5. Prompt generation - Creates context-aware prompts incorporating conversation history
  6. LLM generation - Generates responses using Ollama with appropriate model and parameters
  7. Response filtering - Extracts answer from reasoning traces for deepseek-r1 model
Quest processes everything on your local machine. You’ll need to pull the required Ollama models before using the engine.

​
Architecture

The core components of Quest include:
  • LeetCodeRetriever (retriever2.py:26) - Manages FAISS HNSW index and metadata for 1800+ solutions
  • RAGEngine (rag_engine3.py:18) - Orchestrates retrieval, prompt generation, and LLM calls
  • ConversationHistory (memory_buffer.py:5) - Maintains conversation context with configurable history limit
  • PromptTemplates - Provides model-specific prompt templates for general and reasoning modes
  • Flask API (app.py:9) - REST API with endpoints for search, mode switching, and history management

​
Dataset

Quest includes a pre-generated dataset of 1800+ LeetCode solutions in JSON format, created using the qwen2.5-coder:1.5b model. Each solution includes:
  • Complete code implementation
  • Problem difficulty level
  • Related topics and concepts
  • Companies that ask the problem
  • Approach explanations
The solutions dataset is compressed in leetcode_solutions.zip. You’ll need to unzip it during installation.

​
Next steps

Ready to get started? Follow our installation guide to set up Quest on your machine.

Install Quest

Install dependencies, unzip the dataset, and verify your setup

Build docs developers (and LLMs) love

Get started for freeTalk to us