feat: added readme metadata for hf spaces

update

refactor: keep content and fact as metadata while using combined text for embeddings

refactor: combine content and fact into text column with max length 8192

feat: add milvus token and increase field max lengths

fix: truncate fact and conclusion to 2048 chars in loader

refactor: clean up imports and formatting in data loader

refactor: change references from list to string type

feat: add verbose flag to log llm prompts

feat: add references metadata to hoax detection system

feat: update verdict labels to indonesian in llm prompt

feat: migrate to pydantic v2 and add requirements.txt

feat: add data loading functionality with vector store integration

feat: add batch data insertion to vector store

feat: enhance api with batch processing and better docs

refactor: improve cli, api and gradio integration

feat: add core module structure with config, models and services

refactor: move trusted domains config to hoax_detect/config

perf: batch insert embeddings into milvus in chunks of 1000

style: format embeddings.py with consistent spacing and comments

feat: ensure milvus collection exists before embedding

feat: add references column to embeddings process

refactor: move embedding and insertion logic to embeddings.py

refactor: move trusted domains to json config file

feat: restrict tavily api searches to trusted indonesian news domains

feat: enhance llm prompt to consider both sources and own knowledge

(no commit message provided)

feat: add basic gradio web app to call fact check api

feat: add fastapi-based fact checking api with verdict, explanation, sources

feat: add similarity threshold filtering to vector search results

fix: increase varchar max_length for content fields in milvus schema

feat: combine vector db and tavily api results for improved fact checking

feat: add tavily-only search mode to bypass vector db retrieval

docs: add tavily api key to env example for news search integration

feat: include sources in prompt for retrieved hoax chunks

fix: update openrouter model to gemini-2.0-flash-001 in api call

feat: append sumber field to text in dataset loader

fix: enforce varchar length limits on all fields before milvus insertion

chore: comment out row limit in dataset loader for full data access

feat: add data_summary.py to report max character length per column

feat: add openrouter llm verdict with --no-llm option to main script

feat: add --skip-embedding argument to bypass embedding and start searching

feat: add milvus vector search and retrieval demo to main program

feat: separate chunk text and hoax content metadata in vector store

fix: batch insert data as list of lists to avoid DataTypeNotSupportException

refactor: simplify embedding insertion by removing intermediate lists

feat: chunk long texts into multiple entries for vector database storage

fix: update insert_embeddings to use pre-concatenated text and adjust truncation

chore: limit dataset loading to first 100 rows temporarily

fix: reduce content and fact field truncation length from 2048 to 2000