Spaces:
Sleeping
Sleeping
Commit History
-
updated
Verified signaturec8c8788 -
Update the Readme
Verified signatureca3481e -
Remove extra markdown docs before HF upload
Mr-GondalKeep README.md as the only Markdown documentation file and remove draft/review/spec notes from the upload-ready project tree.68be2c2 -
Codexx polish constellation interactions
Mr-GondalCodexx removed noisy always-visible paper labels from the constellation canvas while keeping click-to-detail behavior and explicit Open paper links. This also preserves the darker Gradio UI polish, fixes compare dropdown defaults, improves search/result interaction behavior, and adds regression coverage for the constellation renderer.ee4fab6 -
Prepare Scholar Lens for release
Mr-Gondal763e1c2 -
Fix Modal inference startup for Scholar Lens
Mr-Gondal- Deploy Modal backend under a clean v2 app name - Pin transformers to 4.51.3 to fix Qwen2 tokenizer startup crash in vLLM - Keep CUDA devel image for reliable vLLM/flashinfer GPU startup - Restore working AI summarize and synthesis endpoints for the Gradio Space1124e00
-
Fix Modal vLLM CUDA startup
Mr-Gondalf813ac4 -
Add 3D paper constellation and AI paper comparison
Mr-GondalAdds two major Scholar Lens features: - Adds a Constellation tab with a 3D interactive paper map built from OpenAlex/search-result data - Exports constellation corpus data as a ZIP with graph JSON and data-completeness metadata - Adds keyword fallback networks when citation/co-citation data is sparse - Adds a Compare tab for AI-assisted comparison between any two searched papers - Updates search reset/dropdown wiring for the new workflows - Improves collapsed tab overflow menu readability in the dark theme - Adds tests for author handling, constellation fallback metadata, corpus ZIP export, and reset behavior78a5c66
-
Remove Python cache files
Mr-Gondald5b5c0c -
Merge Hugging Face Space baseline
Mr-Gondal57b68cb -
Add Hugging Face Space attributes
Mr-Gondal6126d41 -
feat: add local inference and submission notes
Mr-GondalAdd a llama.cpp/GGUF local inference path, Field Notes report, benchmark documentation, BibTeX export, row-selection reliability improvements, and configurable Modal warm windows for demos.f724565 -
docs: map app to Backyard rubric
Mr-GondalAdd judge-facing proof points in the About tab and README so the submission clearly addresses problem specificity, adoption evidence, small-model fit, and polish. Cover the rendered panel with a unit test.f3d5594 -
feat: align app with small-model strategy
Mr-GondalSwitch the Modal default to Qwen2.5 3B and update the product copy around atmospheric-science literature review. This makes the Ask experience and README match the Build Small/NVIDIA story before later polish work.c88dd33 -
Add paper results context and AI paper chat
Mr-GondalAdd a paper-focused AI workflow for discussing selected or pasted research papers. - Add optional Results / Findings input alongside paper abstract/context - Summarize using both abstract/context and results/findings when available - Add a paper chat prompt and Ask About Paper action - Reset paper chat state when selecting a new paper or clearing the app - Include results/findings in exported summary Markdown - Update Modal summary prompts to prioritize results/findings when present - Add tests for combined paper context and summary export behavior460ce99
-
Finish professional review improvements
Mr-GondalComplete the remaining professional review recommendations across UX polish, data quality, exports, and tests. - Add search insight cards, sample prompts, evidence snippets, and export actions - Add copy/download affordances for summaries and result tables - Improve arXiv/PubMed query construction and fuzzy deduplication - Use an environment variable for the contact email - Return proper HTTP 400 errors for invalid Modal requests - Improve Modal error display in the app - Style previously unused UI classes - Expand unit tests for parsing, dedupe, exports, query building, and selection behavior0ee3796 -
Improve search reliability and medium-priority UX
Mr-GondalAddress medium-priority review findings across search quality, pagination, responsiveness, and error handling. - Add retry/backoff for transient academic API failures - Rank search results by query relevance before year and citations - Extract cleaner search terms from natural-language Ask questions - Disable pagination buttons when no previous/next page exists - Make the results table horizontally scroll on small screens - Hide raw Modal generation exceptions from user responses - Add focused unit tests for ranking, query extraction, pagination, context budgets, and clear reset behavior63aa370 -
Add input limits and complete clear reset
Mr-GondalProtect Modal inference from oversized inputs and make the Clear button reset all visible app state. - Add UI max-length limits for search, ask, and summarize inputs - Add rough token counting and app-side input validation - Enforce synthesis context budgets before Modal calls - Add backend Modal input limits before vLLM tokenization - Chunk long summarization inputs and combine section summaries - Reset Ask, Search, Summarize, references, selector, and hidden row state from Clear - Document input protection behavior in READMEac834a3 -
Secure Modal inference endpoints
Mr-GondalMove Modal endpoint configuration out of source code and protect inference calls with a shared bearer token. - Read Modal endpoint URLs from environment variables - Add SCHOLAR_LENS_MODAL_TOKEN for authenticated app-to-Modal calls - Send Authorization header from the Gradio app - Reject unauthorized requests inside Modal endpoints - Document required environment variables and Modal secret setup87cfca0 -
Use Mistral Small 3.1 for judge demo inference
Mr-GondalUpgrade the Modal inference backend from Qwen to Mistral Small 3.1 24B for the final demo. - Replace Qwen model with mistralai/Mistral-Small-3.1-24B-Instruct-2503 - Switch Modal inference from Transformers/bitsandbytes to vLLM - Use A100-80GB for reliable short demo serving - Limit model context to 32k to reduce memory pressure - Shorten Modal warm window to avoid idle GPU spend - Update app and README copy to match the new modeld83ce62 -
Wire search result row selection
Mr-GondalMake search result rows actually selectable from the visible results table. - Add row IDs, click handlers, keyboard handlers, and accessibility attributes to result rows - Prevent paper links from triggering row selection - Add hidden Gradio bridge inputs/buttons for row-click selection - Reuse the existing summarize flow when a row is clicked - Keep the dropdown selection synced with the clicked resultb5b26ec -
feat: add grounded Q&A and faster multi-source search
Mr-GondalCodex-assisted changes: - Add grounded Q&A flow with reliable paper selection - Replace Semantic Scholar with OpenAlex and Crossref - Keep arXiv and PubMed for 4-source parallel search - Reconstruct OpenAlex abstracts and strip Crossref JATS - Add DOI-aware deduplication with title fallback - Paginate results table with Prev/Next controls - Improve Modal inference startup performancedf2b7ae -
Fix inference startup and deduplication with Codex-assisted cleanup
Mr-Gondale95bb2d
-
[codex] Session 2: Improve paper selection and AI summarization flow
Mr-Gondal8e2ff16 -
[codex] Added Modal AI summarization integration
Mr-Gondale559d16 -
[codex] Session 1: Core Gradio app with search across Semantic Scholar, arXiv, and PubMed
Mr-Gondal744c015 -
Initial commit
Haris HussainUnverified signature45f3061 -
initial commit
Verified signature36b54e2