✳Cheng Chen / Projects
Research software · Search & reading

Research starts with a better trail.

AutoSearch AI brings web search, academic discovery, citation-aware synthesis, and an arXiv paper radar into one reading workflow—so sources remain visible as the answer takes shape.

View source & setup ↗Explore the pipeline ↓
Illustrative search · backend not running here
⌕   Efficient multimodal models · arXiv mode
01Find relevant papers and source metadata
02Extract evidence before synthesizing an answer
03Follow citations into a focused PDF reading view
Project previewThis page introduces the open-source system. Live research search needs its FastAPI backend, search providers, and an LLM configuration; GitHub Pages serves only this static preview.
Inside the system

Keep the evidence in view

The repository contains a Python/FastAPI pipeline and a Next.js reading interface—not merely a search-box mockup.

01 / DISCOVER

Search across modes

Switch among quick, deep, academic, and arXiv modes according to the question.

02 / SYNTHESIZE

Ground the answer

Aggregate sources, extract content, and stream an answer with citation links and follow-up queries.

03 / READ

Go back to the paper

Use paper metadata, method notes, and the integrated reader to inspect the original PDF.

Run the full system

A local research stack, with visible sources.

The search API cannot execute on GitHub Pages. The repository documents local development and Docker Compose, including optional search engines, model settings, and runtime LLM overrides.

Open setup guide ↗
  • BackendPython, FastAPI, search aggregation, content extraction
  • FrontendNext.js search interface and PDF reader
  • ModesQuick, deep, academic, and arXiv paper radar
  • Deployment boundaryLive queries need a running backend; no API credentials are published here.