Deep Dive: Olowo Omondi Philly - Backend Development & the TTS Pipeline
How Olowo Omondi Philly built the FastAPI backend that powers the URA Chatbot and engineered the text-to-speech pipeline that gives it a voice.

Olowo Omondi Philly
Deep Dive: Olowo Omondi Philly - Backend Development & the TTS Pipeline
Introduction
Olowo Omondi Philly built the backbone of the URA Chatbot: the FastAPI service that ties the frontend, the retrieval engine, and the model together, plus the text-to-speech pipeline that lets the chatbot answer out loud.
Backend Architecture
Philly designed a clean, layered FastAPI backend:
API Layer
- RESTful endpoints for chat, history, and analytics
- Streaming responses via Server-Sent Events (SSE)
- Request/response validation with Pydantic v2
- Rate limiting and authentication
Service Layer
- Orchestration of the RAG pipeline
- Query processing and response formatting
- Session and conversation handling
Data Layer
- PostgreSQL access with proper indexing and migrations
- Vector store (Qdrant) integration for retrieval
- Redis caching for hot paths
The TTS Pipeline
Many taxpayers prefer to listen rather than read, so Philly built a text-to-speech pipeline for both supported languages:
- Language-aware synthesis: English and Luganda voices
- Phonetic optimization: language-specific pronunciation handling
- Streaming audio: responses spoken back as they are generated
- Fallback handling: graceful degradation to text when audio is unavailable
Reliability
Philly made the backend dependable under load:
- Async request handling for high concurrency
- Connection pooling for the database and vector store
- Structured error handling and health-check endpoints
- Clear API contracts documented with OpenAPI/Swagger
Impact
Philly's backend and TTS work delivered:
- A service tested to handle 1000+ concurrent users
- Sub-2-second average response times
- Spoken answers in English and Luganda for accessibility
- A stable, well-documented API the whole team built against