CodeMap.AI

• 

  Welcome Screen

• 

  RAG Architecture

• 

  Example Prompt

• 

  Example Follow Up

Inspiration

The inspiration behind CodeMap.AI was to create a tool that could serve as a personalized guide for students and professionals navigating the complexities of coding. We recognized the challenges faced by learners at all levels—whether they are beginners just starting out or advanced coders looking to delve into specialized domains like machine learning or blockchain. We wanted to create a platform that not only provides expert guidance but also tailors its responses to each user's unique journey, making learning more efficient and accessible.

What it does

• CodeMap.AI is a web application providing technical guidance through Cody, an AI assistant.
• Cody uses advanced Retrieval-Augmented Generation (RAG) technology and the LLaMa 3.1 400B base model.
• It relies on a vectorized knowledge base of over 1,000 curated documents.
• Cody offers accurate, context-aware advice across various technical domains including web development, machine learning, blockchain, and cybersecurity.
• It provides personalized learning pathways, expert guidance, and support for independent learning based on user queries and conversation history.

How we built it

• Developed using the RAG architecture combining retrieval-based and generation-based techniques.
• Backend powered by a TiDB vector database for efficient document retrieval.
• Used the LangChain library for handling document vector embeddings.
• Deployed on Streamlit with a user-friendly chat interface for interacting with Cody.
• All resources used are free and open source, allowing for reproducibility and utilization by individuals without any paywall.

Challenges we ran into

• Integrating RAG architecture with a vast and diverse knowledge base to ensure precise, contextually relevant responses.
• Optimizing performance to handle complex queries efficiently given the large volume of data.
• Designing an intuitive user interface on Streamlit that accommodates Cody's dynamic interactions.
• Implementing a context window for maintaining conversation relevance.
• Using a custom-hosted LLM with LangChain and the LLM abstract class for advanced functionality.

Accomplishments that we're proud of

• Successfully developed and deployed CodeMap.AI as a functional MVP.
• Cody provides expert-level guidance across multiple technical domains and tailors advice based on user interactions.
• Seamless integration of the TiDB vector database and LangChain library, enhancing efficiency and accuracy.
• Solved challenges related to context window implementation and custom-hosted LLM integration, contributing to a robust RAG AI system.
• Ensured that all used resources are free and open source, making the software accessible for reproduction and personal use without financial barriers.

What we learned

• Balancing complexity with usability is crucial for delivering accurate and relevant information effectively.
• Managing and curating a large knowledge base and ensuring efficient data retrieval are key to enhancing the user experience.
• Implementing context windows, RAG architecture, and custom-hosted LLMs requires careful consideration of both technical and user experience aspects to ensure the system operates efficiently and provides relevant responses.

What's next for CodeMap.AI

• Expanding the knowledge base to include more advanced topics and emerging technologies.
• Integrating additional technical domains and industry-specific knowledge.
• Enhancing personalization with advanced algorithms and developing interactive coding exercises.
• Exploring integrations with popular development tools.
• Expanding language support to cater to a global audience with localized technical content.

Built With

• custom-apis
• langchain
• llama-3.1-400b
• llm
• python
• rag
• streamlit
• tidb-vector-database