Pinned Loading
-
evalhawk/evalhawk
evalhawk/evalhawk PublicTrustworthy LLM evaluation: calibrate your judges against humans and get bias-corrected results with honest error bars.
Python 1
-
graph-chunker
graph-chunker PublicStructure-aware PDF chunking for RAG. It builds a document hierarchy from multiple PDFs and splits chunks along section boundaries. Output is a NetworkX graph linking sections, tables, figures, cro…
Python
-
hybrid-autocomplete
hybrid-autocomplete PublicSearch-as-you-type autocomplete that pairs a fast trie (prefix, phrase, partial-word and typo matching) with a semantic fallback for synonyms. Includes a small Flask API and a minimal demo page. Em…
Python
-
mcq-studio
mcq-studio PublicGenerate source-grounded multiple-choice quizzes from PDFs with LangGraph. Every question is schema-validated, reviewed by an LLM judge, and cited to its source page. Works with Hugging Face, OpenA…
Python
-
walmart-eda-analysis
walmart-eda-analysis PublicExploratory Data Analysis and ARIMA time series forecasting for Walmart sales data. Modular Python package with comprehensive tests and GitHub Actions CI/CD.
Jupyter Notebook
If the problem persists, check the GitHub status page or contact support.
