Movies Recommender

Project Overview

Movies Recommender is an intelligent, LLM-powered movie recommendation system that leverages advanced machine learning techniques to provide personalized movie suggestions through a natural language interface. Datasets tmdb_5000_credits and tmdb_5000_movies

Technical Architecture

Components

• Frontend: Next.js React Application
• Backend: Python FastAPI Service
• Database: PostgreSQL with pgvector
• Observability: LangFuse

Key Technologies

• Frontend: TypeScript, Next.js, React
• Backend: Python, FastAPI, SQLAlchemy, LangChain, LangFuse
• LLM: LLM Models
• Database: PostgreSQL, pgvector
• Authentication: OAuth 2.0

Getting Started

Prerequisites

• Docker
• Docker Compose
• Make

Installation

1. Clone the repository

git clone https://github.com/yourusername/movies-recommender.git
cd movies-recommender

2. Start the application

make up

Preview application

The system can be modeled as a recommendation funnel with several key components that can be represented mathematically:

User Interface Flow Let's represent this as a commutative diagram:

$$ \begin{CD} \text{User-Input} @>>> \text{Embbeding Token} @>>> \text{Query-Vector} \\ @VVV @VVV @VVV \\ \text{Context-Memory} @>>> \text{Recommendation-Engine} @>>> \text{Movie-Suggestions} \\ @VVV @VVV @VVV \\ \text{Evaluation} @>>> \text{Collect-Data} @>>> \text{Model-Update} \end{CD} $$

System Architecture Components The relationship between components can be expressed as:

$$ \text{System} = {F, B, D, L} $$

Where:

• $F$: Frontend (Next.js)
• $B$: Backend (FastAPI)
• $D$: Database (PostgreSQL)
• $L$: LLM Components

Mathematical Structure Analysis The timeline can be represented as a directed acyclic graph (DAG) with the following properties:

$$ \begin{aligned} \text{Timeline} &= (V, E, T) \\ \text{where:} & \\ V &= \text{Set of spans/operations} \\ E &= \text{Dependencies between operations} \\ T &= \text{Temporal mapping function} \end{aligned} $$

Key Components and Relationships Let's create a commutative diagram showing the flow of operations:

$$ \begin{CD} \text{Message} @>>> \text{Process-User-Message} @>>> \text{AskLLM} \\ @VVV @VVV @VVV \\ \text{Movie-Recommendation} @>>> \text{Initiate-Conversation} @>>> \text{Retrieve-Context} \\ @VVV @VVV @VVV \\ \text{Generate-Response} @>>> \text{Store-Assistant-Message} @>>> \text{Evaluation-Pipeline} \end{CD} $$

Key Observations

1. The system implements a hierarchical execution model
2. The trace exhibits nested parallelism in some operations
3. The system follows a pipeline architecture pattern

This trace visualization provides valuable insights into the execution flow and performance characteristics of what appears to be a movie recommendation system built with LangChain.

Development

Project Structure

movies-recommender/
├── recommender-be/   # Backend Python Service
├── recommender-ui/   # Frontend Next.js Application
└── .context/         # Project Context Documentation

Documentation

Detailed project documentation is available in the .context directory, including:

• Project overview
• Development guidelines
• Architecture diagrams

Contributing

1. Read the development guidelines in .context/docs.md
2. Follow the coding standards
3. Submit pull requests with clear descriptions