Smart Office Assistant — Local Chatbot

Local chatbot for corporate environments. A Flask middleware performs linguistic analysis (speech acts, intent, slots, anaphora, style) and routes requests to a Knowledge Graph (Blazegraph/SPARQL) or a Relational Database (PostgreSQL, read-only). Final response generation is handled by OpenHermes (via KoboldCPP). Interface: SillyTavern.

---

Structure

chatbot-middleware/
├─ app.py               · Flask middleware + SSE
├─ config.py            · Settings (endpoints/ports/env)
├─ core/                · Intent & routing (router.py)
├─ features/            · speech_acts, sentiment, context, repairs, style
├─ backends/            · llm_client, kg_client, db_client, sparql_mapper
├─ tests/               · demo_linguistics.py, unit tests
├─ setup_extras/        · helpers (create_db.py, seed_data.py, athens_kg.py)
└─ .gitignore           · Environment and cache protection

---

Quick Start

Requirements: Python 3.10+, KoboldCPP, Blazegraph, PostgreSQL (read-only role)

1. Install dependencies

   pip install -r requirements.txt

2. Service Setup (Example ports)

   • KoboldCPP: 5001 (OpenHermes), 5002 (SPARQL LLM), 5003 (Text-to-SQL LLM)
   • Blazegraph: local SPARQL endpoint
   • PostgreSQL: SELECT only with read-only user
   • Configuration via chatbot-middleware.config.py or environment variables

3. Execution (run commands)

   python -m chatbot-middleware.app

   python -m chatbot-middleware.tests.demo_linguistics

---

Environment (.env examples)

LLM_MAIN_URL=http://localhost:5001
LLM_SPARQL_URL=http://localhost:5002
LLM_SQL_URL=http://localhost:5003
FLASK_HOST=0.0.0.0
FLASK_PORT=5100
BLAZEGRAPH_URL=http://localhost:9999/blazegraph/namespace/kb/sparql
POSTGRES_DSN=postgresql://readonly@localhost:5432/corpdb

The demo (tests/demo_linguistics.py) runs without requiring Blazegraph/PostgreSQL/LLMs.

---

Example Queries

• “Italian restaurant in Plaka, open now.” → KG/SPARQL
• “Show my tasks for today.” → DB/SQL (read-only)
• “Another one there, but cheaper.” → reference + re-ranking (cheaper)
• “List upcoming appointments for Patricia Grant after 15:00.” → DB/SQL
• “Find quiet cafés with Wi-Fi near Syntagma.” → KG/SPARQL

---

License

This project is licensed under the MIT License.