Benjamin Agyekum

About

Backend/full-stack TypeScript engineer who has owned end-to-end, production-oriented systems including an AI property management platform (NestJS/Postgres/WebSockets on Google Cloud using Gemini Vision) and an AI logistics platform (Node/Redis queues/Postgres) focused on low-latency, correctness, and observability. Also designed a public GraphQL API and TypeScript SDK for education partners at StudyFetch, citing 40+ partner integrations in the first quarter.

Key Strengths

• Owned end-to-end backend architecture and implementation (NestJS/Postgres/WebSockets) for an AI property management platform
• Designed event-driven, real-time workflows using AI agents to classify requests and trigger dispatch automation
• Built production-oriented reliability practices: strict TypeScript, CI/CD with unit/integration tests, structured logging and Prometheus metrics
• Scaled low-latency backend for real-time logistics using Redis queues and Postgres with idempotency and exponential backoff retries
• Designed and shipped a public GraphQL API + TypeScript SDK with semantic versioning, deprecation strategy, and strong developer documentation
• Demonstrated external developer adoption impact (reported 40+ partner integrations in first quarter at StudyFetch)
• Hands-on browser automation for checkout/E2E flows with Playwright/Puppeteer; uses resilient locators and mitigations for flakiness and anti-bot detection
• Shipped GPT-4 tutoring agent converting lecture transcripts/notes/PDFs/slides into interactive tutoring sessions
• Designed production reliability for LLM agents (99.9% uptime target, robust error handling, graceful degradation)
• Built monitoring for latency, token usage, and response-quality drift over time
• Implemented RAG pipeline with learning-objective identification, vector retrieval, adaptive responses, and conversational memory
• Established offline + online evaluation loop (synthetic CI evals, telemetry, dashboards, log-based error analysis)
• Improved reliability by adding a disambiguation step to reduce schema-validation retries from ambiguous user inputs
• Architected agentic routing system with structured intent parsing, tool/function calling, and state management in NestJS + LangChain
• Hardened agent predictability with JSON schema enforcement, automatic retries, and state-transition guards
• Integrated messy multi-vendor logistics data using validation/normalization, dynamic schema registry, and idempotent operation IDs
• Reported business impact: ~23–30% improvement in student comprehension/retention; ~60% reduction in agent errors in logistics integration

Experience

Education

Skills

Languages

Certifications