PC Part Picker LLM — Fine-Tuned PC Build Generator
This project is a learning experiment designed to explore the full pipeline of fine-tuning a large language model (LLM) and performing inference on a structured, domain-specific tasks in this case, generating tailored PC builds.
Overview
This model was created as a learning project to explore:
Fine-tuning of large language models (LLMs)
Domain-specific structured data inference
Dataset preparation and prompt design
Model evaluation and output testing
I scraped approximately 6,000 real-world PC builds from PCPartPicker using my custom Selenium-based Python bot PC-Part-Picker-Scrapper-to-JSON. These included diverse combinations of CPUs, GPUs, RAM, storage, and other components. The builds served as a dataset for training a transformer-based language model to learn compatibility patterns and generate new builds.
The primary objective was to evaluate whether an LLM could understand component relationships and generate coherent, purpose driven PC builds in response to natural language prompts such as:
"Build me a $1500 gaming PC for 1440p ultra settings."
"I need a workstation PC for video editing under $2000."
Features
Generates complete PC builds tailored to user requirements
Supports natural language input for task-specific generation
Outputs include the following components:
CPU
GPU
Motherboard
RAM
Storage
Power supply
Case
Cooling (when required)
Example Usage
Prompt:
Build a $1000 PC for light gaming and everyday tasks.
