StandupSync Documentation Assistant

Transform daily standups into living documentation through real-time AI transcription and intelligent document updates.

What This Skill Does

This skill helps you build and maintain StandupSync, an AI-native documentation assistant that:

Records audio standups with real-time GPT-4o transcription

Uses Voyage AI embeddings to find relevant documents

Intelligently updates documentation with standup content via Qdrant vector search

Streams live updates using Vercel AI SDK

Manages documents in Supabase with RLS security

Tech Stack

**Frontend:** Next.js 15 (App Router), TypeScript, Tailwind CSS, shadcn/ui

**AI/ML:** OpenAI GPT-4o, Instructor (structured outputs), Voyage AI embeddings

**Data:** Supabase (PostgreSQL + auth), Qdrant Cloud (vector search)

**Deployment:** Vercel

Project Structure

```

/app

/api

/transcribe # GPT-4o audio transcription

/analyze # Process transcript chunks

/docs # Document CRUD

/dashboard # Main UI

/docs # Document viewer/editor

/components

/ui # shadcn components

/audio # Recording controls

/documents # Doc viewer, update indicators

/lib

/openai # GPT-4o utilities

/ai # Instructor integration

/embeddings # Voyage AI

/vectordb # Qdrant client

/supabase # DB client

```

Development Instructions

1. Environment Setup

Create `.env.local` with:

```

NEXT_PUBLIC_SUPABASE_URL=

NEXT_PUBLIC_SUPABASE_ANON_KEY=

SUPABASE_SERVICE_ROLE_KEY=

OPENAI_API_KEY=

VOYAGE_API_KEY=

QDRANT_URL=

QDRANT_API_KEY=

```

2. Code Patterns

**Server Components by Default**

Use `'use client'` only for interactivity

Keep all AI logic server-side

Stream responses with Vercel AI SDK

**Type Safety**

Define TypeScript interfaces for all API responses

Use Zod for runtime validation

Use Instructor for structured LLM outputs

**Error Handling**

```typescript

export async function POST(req: Request) {

try {

const body = await req.json()

// Validate with Zod

// Process

return Response.json({ success: true, data })

} catch (error) {

console.error('Detailed error:', error)

return Response.json(

{ error: 'User-friendly message' },

{ status: 500 }

)

}

```

3. Database Schema

**Supabase Tables:**

`documents` - id, user_id, title, content (markdown), created_at, updated_at

`standups` - id, user_id, transcript, audio_url, created_at

`doc_updates` - id, standup_id, document_id, section, new_content, created_at

**Enable RLS on all tables** - Users can only access their own data.

4. Styling Conventions

Tailwind CSS only (no custom CSS)

Mobile-first responsive design

Use shadcn/ui dark mode variants

Consistent spacing with Tailwind scale (4, 8, 12, 16)

5. AI Generation Rules

Use Instructor for structured outputs (never parse raw text)

Stream responses for operations >2 seconds

Graceful fallbacks if AI fails

Use embeddings to filter relevant context before LLM calls

6. Component Conventions

PascalCase components, kebab-case files

One component per file

TypeScript interfaces for all props

Extract common patterns to `/components/ui`

7. Adding New Features

1. Define types/interfaces first

2. Build API route with validation

3. Create UI component

4. Test each phase independently

5. Commit after each working feature

6. Use descriptive commit messages: `feat: add live transcription display`

8. Testing Checklist

Before demo:

Record → transcribe → doc update (happy path)

API failure handling (user-friendly errors)

Empty states (no docs, no standups)

Loading states (spinners during async ops)

Mobile responsiveness

9. Demo Flow

1. Show empty state

2. Create sample document ("Sprint 1 Status")

3. Click "Record Standup"

4. Speak: "Yesterday I fixed the Redis timeout bug. Today I'm working on the new API endpoint. No blockers."

5. Show live transcription appearing

6. Show AI analyzing and updating document

7. Show seamless content insertion

8. Stop recording

9. Show final updated document

Important Constraints

**DO NOT** change files unless explicitly asked

**DO NOT** add unrequested features

**DO NOT** use external CSS files

**DO NOT** hardcode API keys

**DO NOT** skip error handling

**NEVER** commit `.env.local` or `node_modules`

**ASK** if user intent is unclear

Security Requirements

All secrets in `.env.local` only

Server-side validation for all inputs

Supabase RLS for data access control

Rate limiting on API routes

Never expose API keys or internal errors to users

Notes

This is designed as a hackathon-speed project (3-hour build window). Prioritize working features over perfect code. Focus on the demo flow showing real-time transcription → AI analysis → live document updates.

StandupSync Documentation Assistant

StandupSync Documentation Assistant

What This Skill Does

Tech Stack

Project Structure

Development Instructions

1. Environment Setup

2. Code Patterns

3. Database Schema

4. Styling Conventions

5. AI Generation Rules

6. Component Conventions

7. Adding New Features

8. Testing Checklist

9. Demo Flow

Important Constraints

Security Requirements

Notes

Reviews (0)