Semantic Search Engine UI

This repository provides a modern web interface for the semantic‑search‑engine, built with FastAPI and Jinja2 templates.

It provides data preparation, collection and document management, semantic vector search, hybrid search, generative question answering (RAG) with user rating, and search history.

Getting Started

Installation and Launch

cd fastapi_ui
python3 -m venv .venv
.venv/bin/pip install -r requirements.txt
bash run.sh                     # launches on http://localhost:8000

Alternatively, you can launch the application directly with Uvicorn:

cd fastapi_ui
uvicorn server:app --host 0.0.0.0 --port 8000

Environment Variables

Variable	Default	Description
UI_HOST	0.0.0.0	Uvicorn host / listen address
UI_PORT	8000	Uvicorn port
UI_CONFIG_PATH	configs/ui-configuration.json	Path to API backend configuration file
UI_SECRET_KEY	change-me-in-production	Secret key used to sign session cookies
UI_MAX_UPLOAD_SIZE_MB	6000	Maximum size per uploaded file (in MB)
UI_MAX_UPLOAD_FILES	500	Maximum number of files in a single upload request
UI_DEFAULT_LANGUAGE	en	Fallback interface language (en or pl)

Backend Configuration

Backend endpoint URLs and hosts are configured in configs/ui-configuration.json (or the path set via UI_CONFIG_PATH). Endpoints added later (api/indexing_status, api/search_history, …) fall back to their default path when the configuration file does not list them, so an old file keeps working.

Key Features

• Knowledge Base Management:
  • Create, view, and delete collections. A new collection may declare its chunking (max_tokens_in_chunk, number_of_overlap_tokens, prefilled with the engine chunking_defaults) and strict_chunking; the chunk size is capped by max_model_length of the chosen embedder (models_info). Collection cards show the declared chunking and a strict badge.
  • Manage documents, toggle document visibility in search, and inspect/delete text chunks.
  • Upload multiple file formats (.pdf, .docx, .txt, .html, .md, .zip) or paste raw text, split with the settings of the collection or with custom chunking; the engine errors CHUNKING_TOO_LONG (000005_DATA) and CHUNKING_MISMATCH (000006_DATA) are shown inline on the form.
  • Indexing progress (GET api/indexing_status): a panel shows how many documents are indexed, which one the worker is embedding, how many are still waiting and which ones failed; document rows carry a status badge. Indexing runs inside the request or in a Celery + RabbitMQ queue on the engine side (ENV_USE_CELERY=1) — the panel looks the same, it just moves on its own.
• Semantic & Hybrid Search:
  • Vector similarity search with cosine similarity scoring.
  • Hybrid search (RRF) combining vector search and full-text keyword retrieval.
  • Result reranking with dedicated reranker models.
  • Generative QA: asynchronous answer generation based on retrieved passages, with multi-star rating and user feedback.
  • Search history per collection with pre-filled filters and parameters.
• Modern UI & UX:
  • Lightweight, responsive design without external CDN dependencies (works fully offline).
  • Dark / Light / Auto theme support.
  • Bilingual interface (Polish and English) with runtime locale switching (EN | PL).
  • OpenAPI documentation available at /docs and health check at /healthz.

Testing

Smoke tests run against a local mock backend (tests/mock_sse_backend.py) and validate all views, authentication, search flows, and localization:

cd fastapi_ui
.venv/bin/pip install httpx
.venv/bin/python tests/smoke_test.py

For more details on application architecture, endpoints, and components, see fastapi_ui/README.md.