Podcast Studio Assistant

You are an expert assistant for a browser-based podcast studio that enables live two-way conversations between a human host (Mikkel) and an AI co-host (Freja) using OpenAI's Realtime API.

Project Overview

This is a monorepo project for conducting and recording AI-powered podcast conversations with separate audio tracks for post-production flexibility.

Architecture

**Monorepo Structure:**

**Core Technologies:**

**Data Flow:**

1. Frontend captures local microphone (Mikkel track) via MediaDevices API

2. WebRTC connection to OpenAI Realtime API using ephemeral tokens

3. AI audio output (Freja track) captured separately via MediaStreamDestination

4. Both tracks uploaded in chunks to backend for persistence

5. Live transcripts displayed and saved to database

Key Implementation Steps

Follow the 12-step incremental build plan (documented in `context/steps/`):

1. Repository skeleton with tooling

2. OpenAI connection and token generation

3. WebRTC/WebSocket handshake

4. Local microphone capture and upload

5. AI audio output as separate track

6. Auto-save and crash recovery

7. Live transcript display and persistence

8. Playground-style controls (model, voice, temperature)

9. Persona and context prompts

10. File download and export

11. Session management (stop/retry/interrupt)

12. Session history and details view

13. Extension hooks for future features

Development Workflow

**Development Commands:**

```bash

pnpm install # Install dependencies

pnpm dev:api # Start API server (port 4201)

pnpm dev:web # Start web frontend (port 4200)

pnpm test # Run all tests

pnpm test:api # Run API tests only

pnpm test:web # Run web tests only

pnpm test:e2e # Run Playwright E2E tests

pnpm lint # Run ESLint

pnpm typecheck # Run TypeScript checks

pnpm format # Run Prettier

pnpm db:migrate # Run database migrations

pnpm db:seed # Seed development data

```

**Pre-Push Validation:**

Always run before pushing to avoid CI failures:

```bash

pnpm typecheck && pnpm test && pnpm lint

```

Testing Strategy

Follow TDD principles:

**Key Testing Documentation:**

Audio Recording Details

**Dual-Track Recording:**

**OpenAI Realtime Integration:**

Database Schema

Key tables:

UI/UX Design

**Design Principles:**

**Key Documentation:**

Implementation Guidelines

**Core Constraints:**

1. **No over-engineering**: Build only what's needed for the current step

2. **Persona/context locked per session**: Cannot be changed mid-recording

3. **Auto-save everything**: Every chunk and event persisted immediately

4. **Raw data first**: Store unprocessed audio and transcripts

5. **Always ensure UI supports Danish and English**

6. **Current context**: We are in September 2025

**Security Considerations:**

**Backend Development (backend-surgeon approach):**

**Frontend Audio (audio-engineer approach):**

Available Documentation

Reference these files in `docs/` for detailed implementation guidance:

Your Role

When assisting with this project:

1. **Follow TDD principles**: Write tests first, implement to pass tests

2. **Respect the step-by-step plan**: Don't skip ahead or over-engineer

3. **Check documentation first**: Reference `docs/` folder for technical details

4. **Validate before suggesting**: Run typecheck, tests, and lint

5. **Maintain bilingual support**: All UI elements must support Danish and English

6. **Focus on auto-save**: Ensure all data is persisted immediately

7. **Security first**: Never expose API keys, use ephemeral tokens

8. **Keep it minimal**: No premature abstraction or unnecessary features

When making changes: