Wikipedia Offline ADV — Standalone search engine and Pseudo-AI assistant containing 6,400,233 articles for the M5Stack Cardputer ADV.
A project for Chinese Extractive Question Answering (QA), focusing on selecting relevant text spans from a given context to answer specific questions. It supports a two-step process for paragraph and span selection or an end-to-end solution using fine-tuned models like BERT.
Extractive QA pipeline — no LLM, no training. BM25 + FAISS + Wikidata + cross-encoder re-ranking + roberta-base-squad2.
AI-powered Question Answering app using Gradio and Transformers
An extractive Question Answering system built with Hugging Face Transformers and Streamlit. Fine-tuned a DistilBERT model on the SQuAD dataset and deployed it as an interactive web app for real-time QA from user-provided context.
Ask questions about your own documents with a fine-tuned BERT model. Extractive QA with train / eval / ask CLI.
Local provenance-first document processing with OCR, BM25 retrieval, evidence-grounded extractive QA, and explicit abstention.
Local-first RAG engine built on the Laya decision engine: hybrid BM25 + dense recall, calibrated relevance re-ranking and answerability gating that return evidence with probabilities rather than generated prose. CLI, FastAPI HTTP API, MCP server and web GUI.
Fine-tuning MuRIL for extractive question answering in Telugu elementary mathematics. 57.7 F1 on genuine comprehension against a 2.7% rule-based baseline — plus evidence that a 100.0 F1 on templated data measures copying, not reasoning.
To associate your repository with the extractive-qa topic, visit your repo's landing page and select "manage topics."