Project Overview

I led the development of a suite of AI-powered APIs designed to give developers secure, scalable access to our internal AI capabilities. This was a greenfield opportunity to build a modern API platform from scratch, enabling use cases such as document reading, validation, and verification. The result was a unified developer experience, capable of powering AI integrations across real estate, insurance, and operations workflows.

Technical Overview

We designed our API platform with performance, flexibility, and maintainability in mind. I led the design and implementation of the entire backend architecture using a modern stack.

Open API

I built the API platform that powered the new AI capabilities, some of the tech choices included:

• I built the OpenAPI 3.0-compliant server using Hono, a blazing-fast web framework.
• I defined endpoint contracts, enforced strict type safety with TypeScript
• I leveraged Vitepress for interactive documentation, including markdown support and autogenerated OpenAPI references.
• The system was backed by Postgres via Drizzle ORM
• It was supported by BullMQ for distributed job queues, automatic retry mechanisms, and other scaling optimizations.
• The system was cached with Redis.
• Testing and CI were powered by Vitest, enabling fast, reliable test cycles across services.

Vision AI API

The Vision AI API extracts structured data from uploaded documents. It supports:

• Validation: Authenticity, format, and forgery checks
• Classification: Automatic document type detection
• Extraction: Key data and metadata parsing
• Confidence Scoring: Probabilistic confidence thresholds for downstream systems

Verification AI APIs

We built three primary verification modes to handle real-world document authenticity checks:

• Web Crawling Verification - For documents containing source URLs, I built a headless browser system (using Google Cloud Run Jobs) to dynamically load and interact with poorly structured or JS-heavy pages. We used OpenAI’s computer use tools to navigate, click, and extract page data for comparison.
• Automated Email AI Verification - To automate third-party confirmation, I built a document-driven agent system that generates email identities, sends context-aware verification messages, and manages thread responses. I added smart retries, escalation logic, and even built a full internal React email client to support human follow-up.
• API-Based Verification - For known document types with supported APIs, we made secure, programmatic verification calls and matched results against document contents. This allowed for fast, reliable authentication using authoritative sources.

Roles and Responsibilities

As Head of Engineering, I owned the end-to-end technical vision and execution for this platform. I architected the system from the ground up, designed and implemented core APIs, and ensured type-safe, maintainable code using modern TypeScript tooling. I led backend and infrastructure work, including OpenAPI contract design, async job processing, browser-based crawling systems, and AI pipeline integration.

I also built internal tooling to support operational teams, such as an email-based verification agent system and a React-based escalation dashboard. Beyond coding, I mentored engineers, established our testing and deployment standards, and worked closely with product and operations to translate domain-specific problems into scalable technical solutions.

Impact

• We decreased document processing times by ~85% through automatically reading and validating documents.
• Our verification APIs reduced manual review workloads by 90% by automating third-party confirmations.
• The platform enabled rapid development of AI-powered features across multiple business units, including real estate, insurance, and operations.
• We achieved 99.9% uptime and reliability, with robust error handling and monitoring in place.
• The OpenAPI documentation and interactive examples significantly improved developer onboarding and adoption.