Skip to content

Tech Stack: FastAPI, LangGraph, PostgreSQL, PGVector

An AI agent that generates professional marketing content while maintaining brand consistency through semantic memory.

Creating consistent, high-quality marketing content at scale while maintaining brand voice and factual accuracy.

Built a multimodal marketing-content agent that encodes good taste so teams get on-brand, production-ready outputs on the first try.

  • Designed a flexible RAG engine to select and fill templates from thousands of HTML assets, matching by user intent, content relevance, and structural similarity.
  • Shipped one-shot MJML generation that consistently meets professional design standards.
  • Implemented intelligent prompt routing across wireframe sketches, abstract objectives, and detailed PRDs.
  • Achieved 99%+ first-pass MJML/HTML compilation via an auto-linting + diff-edit loop, slashing render failures and manual rework to near-zero
  • Built advanced template version control for safe rollbacks and iterative improvement.
  • Wrapped it all in a simple, async HTTP service for easy adoption.
  • Agent Architecture: LangGraph for multi-step reasoning and tool use
  • Semantic Memory: PGVector for efficient similarity search across brand assets
  • API Design: FastAPI with async endpoints for real-time content generation

See also: Receipt Reimbursal App