Admind

Admind is an ai powered marketing suite that helps understand the context of your company, your markets, your current traction, and generates full video/image ads to accelerate your growth!

Comment

Inspiration

Content creators and marketers face an impossible challenge: creating high-quality, on-brand content for multiple platforms while analyzing performance and staying consistent with brand voice. We watched small businesses struggle to maintain a social media presence, hiring expensive agencies or spending hours creating content that doesn't resonate. We wanted to democratize professional marketing by building an AI that truly understands your brand and creates content that performs.

What it does

AdMind is an intelligent marketing automation platform that turns your website into a full content creation engine:

  1. AI Company Research - Scrapes your website to understand your brand, products, tone, and values, storing knowledge in a RAG-powered vector database
  2. Social Performance Tracking - Monitors your existing content across YouTube, TikTok, Instagram, and Twitter without requiring OAuth access
  3. AI Marketing Strategy - Combines company insights and social analytics to generate comprehensive marketing plans with target markets, content themes, and posting schedules
  4. Professional Content Generation - Creates branded videos and images complete with voiceover scripts, platform-optimized captions, and relevant hashtags
  5. Brand Consistency - Uses RAG retrieval to ensure every piece of content aligns with your brand voice, values, and messaging

How we built it

Frontend: Next.js 15 + React 19 with TypeScript, Tailwind CSS, and Framer Motion for smooth UI/UX

Backend: Express.js API server with background worker for job processing, deployed separately on Railway

Database: Supabase (PostgreSQL) with pgvector extension for semantic search, row-level security for multi-tenancy

AI/ML:

  • Claude Sonnet 4.5 + GPT-4o for text generation
  • OpenAI embeddings (text-embedding-3-small) for RAG
  • Runway ML for video generation
  • Runware for image generation

Social Integration: Custom API clients for TikTok, Instagram, YouTube, and Twitter using RapidAPI services

Key Innovations:

  • RAG system with 1536-dimensional embeddings for company knowledge
  • Chunked web scraping with Cheerio + Puppeteer
  • Platform-specific content optimization (pacing, hashtags, captions)
  • Background job queue for async processing

Challenges we ran into

RAG Implementation: Building a vector similarity search that actually retrieves relevant company context was harder than expected. We spent days tuning chunk sizes, overlap, and similarity thresholds.

Social API Limitations: Many social APIs require OAuth or have strict rate limits. We solved this by using third-party scraping services that don't require user authentication.

Content Quality: Early AI-generated scripts were generic and off-brand. We implemented RAG retrieval and detailed prompts with brand tone, values, and banned terms to ensure quality.

Multi-API Orchestration: Coordinating between web scraping, embedding generation, LLM calls, and media generation required careful job queue design and error handling.

Platform Differences: Each social platform has different content requirements (caption length, hashtag count, video duration). We built platform-specific optimization rules.

Accomplishments that we're proud of

✅ Complete End-to-End Workflow: From website URL to published-ready content in under 2 minutes

✅ True Brand Consistency: RAG-powered generation that actually sounds like your brand

✅ Real Social Analytics: Tracking live performance metrics without requiring user OAuth

✅ Professional Quality: Scripts paced for voiceover (~2.5 words/second), platform-optimized captions, research-backed hashtags

✅ Production-Ready Architecture: Background workers, job queues, error handling, and scalable database design

What we learned

RAG is Game-Changing: Vector embeddings + semantic search unlocks truly personalized AI that understands context beyond simple prompts.

Prompt Engineering Matters: Spending time crafting detailed prompts with examples, constraints, and brand guidelines dramatically improved output quality.

Social APIs Are Complex: Each platform has different data structures and limitations. Abstracting them into unified interfaces was crucial.

Background Jobs Are Essential: Heavy operations (web scraping, video generation) must run async to keep the UI responsive.

Content Optimization is Platform-Specific: What works on TikTok (short, punchy, hashtag-heavy) fails on LinkedIn (professional, detailed, minimal hashtags).

What's next for Admind

🚀 More Platforms: Add LinkedIn, Pinterest, YouTube Shorts, and Snapchat

📊 A/B Testing: Generate multiple variants and track which performs best

🎯 Performance Prediction: Train ML models on historical data to predict engagement before posting

🎬 Advanced Video Editing: Add text overlays, transitions, and music selection

🤝 Team Collaboration: Multi-user workspaces with approval workflows

📅 Scheduled Publishing: Auto-post content directly to social platforms

💰 ROI Tracking: Connect to ad platforms to measure actual conversion rates

🌐 Multi-Language Support: Generate content in 10+ languages while maintaining brand voice

Built With

Share this project:

Updates