feat: complete intelligence upgrades (semantic matching, dynamic prompts, human feedback loops)

feat: expand rule engine with 30+ new vendors (Google Workspace, Deel, Zendesk, Semrush etc)

fix: switch LLM to llama-3.3-70b and fix flagged API to return amount

fix: graceful ML fallback when peft model version incompatible with installed peft

fix: downgrade ML stack to compatible versions (accelerate, huggingface_hub, tokenizers)

fix: pin peft==0.11.1 and transformers==4.40.0 to fix BloomPreTrainedModel import error

feat: add month-end close API with period locking

feat: add universal transactions ledger api

fix: syntax error missing comma in process.py

feat: show AI reasoning in review queue

feat: add drill-down API endpoint for categorized transactions

feat: expand rule engine with advertising, saas, cloud and coworking vendor whitelists

feat: add rule engine tier 1, expand COA to 17 accounts, fix agent tools

feat: complete architectural upgrade and additional updates

fix: use get_or_create_collection to prevent crash on fresh deployments

fix: move backend files to root for HuggingFace

Add application file

initial commit