sugash028/bse-backend

Add temporary rules export endpoint for local Neo4j migration

Add streaming for LLM mode to fix BOM generation timeout

Add per-message edit and delete

Fix project drift on back-reference queries

Clarify rules are non-deterministic hints in system prompt

Add proper error handling before public release

Migrate conversation storage from SQLite to Neo4j

Add consultant filter/boost to similarity search

Include user note in upload vector search query

Detect topic changes via vector search overlap, not word count alone

Replace fragile history-parsing with server-side conversation cache

Fix project pinning for LLM mode and add reset keyword detection

Fix project drift on follow-up queries — pin to previously found projects

Remove rules_compiled.json — rules already loaded into Neo4j

Add article rules as non-deterministic BOM hints via Neo4j

Pass rich KG details to LLM mode so Claude has full project data

Fix upload default: show similar projects, not BOM, unless user asks

Fix system prompt: answer specific question, only show BOM when asked

Include consultant, project type, refrigerant, warranty in KG results and embeddings

Show BOM only when user explicitly asks, not on every query

Fix Neo4j retry (probe before yield) and schema fläktrum->flaktrum enum

Allow null on all nested schema objects to prevent schema validation errors

Update AI system prompt to prevent hallucination and ground answers in context

Fix Neo4j Aura reconnect, extraction retry, and frontend timeout handling

feat: integrate multilingual-e5-small embeddings and implement frontend API routes for chat and document upload

add sqlite conversation storage

feat: add Docker configuration and deployment documentation for backend service

feat: implement frontend chat interface with markdown support, file upload, and authentication flow

feat: implement backend API routes, core configuration, and frontend chat integration for the quote tool

feat: implement backend API architecture and frontend sidebar for conversation management

chore: ignore auto-generated python bytecode files

feat: add RFQ document upload functionality with automated extraction and LLM-assisted chat integration

feat: add extraction service module to handle data processing logic

feat: implement RFQ document upload extraction service and API endpoints

feat: implement backend chat orchestration with Neo4j integration and initialize frontend project structure

implement LLM-powered chat orchestration with Neo4j vector search and front-end toggle

feat: implement AI-powered BOM drafting tool with Neo4j integration and chat interface

initialize Next.js frontend with chat interface and backend data loaders

initialize backend project structure with Neo4j integration and BOM data loading capabilities