README.md

metadata

title: Algorithm Research Deck
emoji: 🔬
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 6.27.0
python_version: '3.12'
app_file: app.py
pinned: false
short_description: Interactive algorithm research from academic papers
license: apache-2.0

Algorithm Research Deck

Turn an algorithm research topic into an interactive, evidence-aware HTML presentation. DeepSeek-V4-Flash first converts Chinese inputs into a precise English technical interpretation and expands them into historical, milestone, current, and evaluation-oriented search queries. Results are balanced across query intents instead of allowing the first query to dominate. The Space searches arXiv, supplements sparse or empty results with OpenAlex, uses Hugging Face Inference Providers for synthesis, previews the result, and provides HTML, JSON, and ZIP downloads.

Users can disable online paper search. In that mode the deck relies on the model's existing knowledge and visibly discloses that papers and recency were not externally verified. Download controls sit in the presentation toolbar: HTML for the standalone interactive deck, JSON for structured research data, and ZIP for both.

Report generation is staged: selected papers are compressed into evidence cards, then the model infers the research type, creates a topic-specific outline, and writes sections in batches of at most four. Each generated section is locked back to its approved outline identity and layout. Timelines, comparisons, processes, equations, paper analysis, and evidence-gap pages are optional rather than globally required. Layout-aware Pydantic validation, malformed-JSON recovery, citation allowlisting, and targeted repair attempts prevent empty slides and unsupported references. If academic search returns no usable result, the run visibly degrades to an exploratory model-knowledge report instead of failing or fabricating papers.

The standalone deck includes an executive-summary view, a generated research map, slide overview navigation, keyboard and touch controls, interactive process nodes, balanced comparison tables, clickable source chips, responsive layouts, and landscape print/PDF styling.

Space configuration

Secret:

• HF_TOKEN: a fine-grained token with permission to call Hugging Face Inference Providers. Repository write permission is neither needed nor recommended.
• OPENALEX_API_KEY: optional OpenAlex API key; recommended for more reliable Standard/Deep searches and higher rate limits

Variables:

• HF_MODEL_ID: defaults to deepseek-ai/DeepSeek-V4-Flash
• HF_PROVIDER: defaults to auto; optionally choose a currently live provider
• CONTACT_EMAIL: contact address included in polite arXiv requests

The token used on your computer to push this repository must be a write token. Never add that write token to Space Secrets.