LLM-Lightweight Optimized for Dockerized Low RAM Cloud Server

Overview

Key Features

• Customizable search keywords for research paper scraping
• Separating the ingestion pipeline of figure/chart and body content to optimize tokens under free-limit usage in Groq and queue operating using Redis
• Storing extractions in vectorized (RAG; ChromaDB)
• Track/Untrack papers to review later in Google Spreadsheet and weekly summary using LLMs (fixed structure prompt; Cross-domain Meta Synthesis, Links, Emerging Signals, Recommendation Deep Dive)
• Customizable research paper sources, eg. Arvix API

Limitation

• Not developed for multi-user scaling
• Not fully format for some mathematic symbols

Application Screenshots