Academic Workflow — Complex Research Tasks

Overview

For complex multi-step tasks (surveys, analyses, paper writing), break them into discrete steps and execute sequentially. This prevents timeouts and allows progress tracking.

When To Use

•User requests a literature survey or review
•User wants benchmark comparison across papers
•User needs end-to-end research workflow (search → analyze → visualize → write)
•Task involves more than 3 tool calls

Strategy: Divide and Conquer

IMPORTANT: For complex tasks, execute ONE step at a time, report progress, then continue.

Example: Paper Survey Workflow

Instead of trying everything at once:

code

❌ Bad: Try to search, analyze, visualize, and write in one go
✅ Good: Execute step by step with checkpoints

Step-by-Step Template

Step 1: Search and Save

bash

# Search papers and save to JSON
paper-search search "your topic" --max 10 --json > /workspace/projects/papers.json
echo "Step 1 complete: Found $(cat /workspace/projects/papers.json | python3 -c 'import json,sys; print(len(json.load(sys.stdin)))') papers"

Step 2: Extract Data to CSV

bash

/home/user/.venv/bin/python3 << 'PYTHON'
import json
import pandas as pd

with open('/workspace/projects/papers.json') as f:
    papers = json.load(f)

data = []
for p in papers:
    data.append({
        'id': p['id'],
        'title': p['title'][:80],
        'authors': ', '.join(p['authors'][:3]),
        'published': p['published'],
        'categories': ', '.join(p['categories'])
    })

df = pd.DataFrame(data)
df.to_csv('/workspace/output/papers.csv', index=False)
print(f"Step 2 complete: Saved {len(df)} papers to CSV")
PYTHON

Step 3: Visualize

bash

/home/user/.venv/bin/python3 << 'PYTHON'
import pandas as pd
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import seaborn as sns

df = pd.read_csv('/workspace/output/papers.csv')

fig, ax = plt.subplots(figsize=(10, 6))
# Your visualization code here
plt.savefig('/workspace/output/analysis.png', dpi=150, bbox_inches='tight')
print("Step 3 complete: Saved visualization")
PYTHON

Step 4: Generate LaTeX

bash

cat > /workspace/projects/survey.tex << 'LATEX'
\documentclass{article}
\usepackage{graphicx}
\begin{document}
\title{Survey Title}
\maketitle
% Content here
\end{document}
LATEX
echo "Step 4 complete: Generated LaTeX"

Step 5: Compile PDF

bash

cd /workspace/projects && pdflatex -interaction=nonstopmode survey.tex
cp survey.pdf /workspace/output/
echo "Step 5 complete: PDF at /workspace/output/survey.pdf"

Progress Reporting

After each step, report:

•What was completed
•Output file locations
•What comes next

Example output:

code

✅ Step 1/5: Found 10 papers on VLA
   → /workspace/projects/papers.json

✅ Step 2/5: Extracted benchmark data
   → /workspace/output/benchmarks.csv

Continuing to Step 3: Visualization...

Common Workflows

Literature Survey

•Search papers (paper-search)
•Extract metadata to CSV
•Analyze trends (pandas)
•Create visualizations (seaborn)
•Write LaTeX survey
•Generate BibTeX
•Compile PDF

Benchmark Comparison

•Search papers with benchmark mentions
•Extract performance metrics
•Create comparison table
•Visualize results
•Write analysis

Replication Study

•Download paper PDF
•Extract methodology
•Implement code
•Run experiments
•Compare results
•Write report

Timeout Prevention

•Break tasks into 2-3 minute chunks
•Save intermediate results to files
•Use --json output for programmatic processing
•Avoid downloading large files mid-workflow

File Organization

code

/workspace/
├── projects/
│   └── my-survey/
│       ├── papers.json      # Raw search results
│       ├── survey.tex       # LaTeX source
│       ├── references.bib   # BibTeX
│       └── figures/         # Generated plots
└── output/
    ├── survey.pdf           # Final PDF
    ├── data.csv             # Extracted data
    └── analysis.png         # Visualizations