ControlMind/CITATION.cff at main · MorningStar0709/ControlMind · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
cff-version: 1.2.0
title: "ControlMind: MinerU-based Scientific Document Intelligence for Sci-Align, Data Agent, and Medical RAG"
message: >-
  If you use this project in your research, please cite this software and the
  associated dataset (see url below).
type: software
authors:
  - name: MorningStar
    email: fanpuji55@outlook.com
license: CC-BY-4.0
repository-code: "https://github.com/MorningStar0709/ControlMind"
url: "https://huggingface.co/datasets/MorningStar0709/control-sci-corpus"
year: 2026
date-released: 2026-05-24
abstract: >-
  ControlSci is a structured cross-modal alignment benchmark and data agent
  system for control science. Track 1: a 500-question 4-dimension evaluation
  benchmark (Concept Recall / Multi-step Reasoning / Condition Sensitivity /
  Open Design, 125 each) built from 362 control-science documents (23 textbooks
  + 339 arXiv papers) via a fully AI-driven pipeline (MinerU parsing → LLM
  question generation → dual-layer arbitration → embedding fast-track), covering
  14 control sub-domains with 28,514 cross-modal chunks and 253K LaTeX formulas.
  Track 2: ControlSci Data Agent — a 15-intent scientific document cross-modal
  corpus agent powered by a tri-track inference engine (API / Ollama / vLLM)
  with 4-path resource scheduling, closed-loop self-correction, MiMo-V2.5 native
  visual auditing, QLoRA fine-tuning, and a data flywheel for automatic
  benchmark expansion. The project demonstrates that a single-developer,
  full-AI-driven pipeline can produce industrial-grade evaluation benchmarks and
  agent systems without human annotation.
keywords:
  - control-science
  - benchmark
  - scialign
  - cross-modal-alignment
  - llm-evaluation
  - structured-corpus
  - mineru
  - data-agent
  - multi-modal
  - qlora
  - ai4s
  - document-intelligence
  - tri-engine-scheduling