Scriptal Platform

Scriptal is a meeting intelligence platform I built from scratch. Sales teams, consultants, and customer success managers upload their call transcripts and the platform turns them into a searchable, analyzable database of conversation intelligence.

The core of the platform is a hybrid search engine that combines pgvector semantic search with lexical keyword matching. When you search for "pricing objections," it doesn't just look for those exact words — it finds semantically similar passages across your entire transcript library using OpenAI embeddings (text-embedding-3-small, 1536 dimensions) stored as native PostgreSQL vectors with HNSW indexing for sub-second retrieval.

Every transcript goes through an automated AI analysis pipeline. When a file is uploaded, it enters a database-backed job queue. AWS Lambda workers pick up jobs, chunk the transcript (800-word segments with 100-word overlap), generate embeddings, and run structured extraction through GPT-4o-mini. The system detects the meeting type — sales call, support ticket, internal meeting, interview, or customer success check-in — and extracts type-specific insights. A sales call gets objections, buying signals, competitor mentions, and deal stage. A support call gets issue classification, resolution status, and customer sentiment. All extracted to structured database tables, not just raw text.

The Lambda architecture self-scales. A primary function runs on a 5-minute EventBridge schedule, checks queue depth, and spawns up to 10 additional Lambda workers when there's a backlog. Each worker processes 5 jobs concurrently. Peak throughput: 50 transcripts analyzing simultaneously.

There's also a conversational AI interface where users can query their entire transcript library. Ask "what do enterprise prospects say about our onboarding?" and it retrieves relevant chunks via hybrid search, then streams a synthesized answer back using Server-Sent Events so tokens appear in real-time.

The platform handles the full SaaS stack: Supabase Auth with Row Level Security, Stripe billing with tiered subscriptions, team workspaces with seat management, custom webhook integrations with HMAC signing, and an OAuth 2.0 system for the companion Chrome extension (TranscriptExporter) to sync transcripts directly from meeting recording tools.