Skip to content

AInnouncer Studio

Project start: June 2025 — Present

Role: Tech Lead / AI Engineer

Status: 🟢 Active Development

Project description

AInnouncer Studio is a comprehensive AI platform for automatic audio content generation for radio stations, built on a professional LLMs and LLOps architecture.

Radio stations need regular, professional audio content: weather forecasts, music announcements, on-air messages, or advertising materials. This process is time-consuming, expensive, difficult to scale, and heavily dependent on people.

The system combines:

  • text generation (LLM),
  • voice synthesis (TTS),
  • audio mixing (broadcast-ready),
  • automation,
  • monitoring and quality control.

The platform was designed as a scalable SaaS, ready for multiple clients and additional modules.

Architecture Overview

AInnouncer Studio is an event-driven + worker-based architecture:

  • Frontend (Next.js) — configuration of modules, prompts, voices, schedules
  • Backend API (FastAPI) — domain logic, routing, validation, orchestration
  • Asynchronous workers (Dramatiq + Redis) — content generation, TTS, mixing, upload
  • LLM Layer — OpenAI GPT-4o / GPT-4o-mini with advanced system prompts
  • LLOps & Observability — Langfuse (traces, spans, cost, quality), prompt versioning, Promptfoo (prompt testing)
  • Data Layer — PostgreSQL (configurations, prompts, voices, schedules), S3-compatible storage (audio)
  • Infrastructure — Docker Compose, CI/CD, Monitoring (Prometheus + Grafana)

Key Modules

Weather Forecast (Production)

  • weather data → LLM text → TTS voice → jingle mix → upload
  • support for multiple locations and languages
  • broadcast schedules

AInnouncer (DJ / Music Announcer)

  • playlist parsing (.mix)
  • batch text generation
  • announcement frequency control
  • audio ready for broadcast automation

Platform Core (LLOps)

  • prompt versioning
  • response quality monitoring
  • LLM cost analysis
  • retry & fallback logic
  • preparation for AI agents

What I did

  1. Designed the complete AI system architecture in production
  2. Built the backend in FastAPI with separation of concerns
  3. Implemented asynchronous pipelines (Dramatiq)
  4. Created the LLM layer with prompt control and validation
  5. Integrated Langfuse for LLM observability and monitoring
  6. Deployed Promptfoo for prompt testing
  7. Built TTS and audio mixing system to radio standards
  8. Designed CI/CD and cloud environment
  9. Prepared the platform for further AI agent development

Skills

  • Python
  • FastAPI
  • OpenAI GPT-4o
  • ElevenLabs TTS
  • Langfuse
  • Promptfoo
  • PostgreSQL
  • Redis
  • Dramatiq
  • Docker
  • DigitalOcean
  • Next.js
  • TypeScript
  • Prometheus
  • Grafana

Results

  • Fully automatic generation of broadcast-ready audio content
  • Stable, production AI architecture (not a demo)
  • Full control over LLM quality and costs
  • System ready to scale as a SaaS product
  • Solid foundation for:
    • AI agents
    • additional modules (ads, traffic, voice branding)
    • international expansion

Sample photos

Traffic module Settings DJ module Scheduler Weather module Modules overview