Hamza Hassan

• Built and shipped an end-to-end LLM pipeline for resume tailoring and cover letter generation, using prompt engineering and RAG to improve relevance for 12,000+ users.
• Designed model inference stack and production hosting on AWS, deploying quantized models and inference endpoints to reduce latency for user-facing features.
• Implemented data collection and cleaning pipelines for prompt evaluation, curating training examples and human feedback loops to improve output quality.
• Applied Prompt Engineering Fine-Tuning (PEFT) workflows with iterative evaluation, documenting prompts and metrics to guide model updates and product iterations.
• Operated CI/CD for model updates and site deployment with Pulumi and GitHub Actions, ensuring reproducible releases and rollback for production LLMs.

• Coordinated delivery of 35+ Model Context Protocol servers, integrating LLMs into production workflows and enabling standardized model context handling across teams.
• Calibrated evaluation rubrics and reinforcement learning environments used for policy tuning, supporting PPO and Direct Parameter Optimization (DPO) style experiments.
• Automated 90% of evaluation-to-deployment steps for model experiments, including dataset generation, metric calculation and MCP trajectory creation to speed iteration.
• Trained RL models on 170+ open-source repositories, applying Transformer architectures and PyTorch training pipelines to improve code reasoning and multi-language completion.
• Built multi-agent experiment orchestration with n8n integrations to reproduce evaluation flows and persist experiment state for model comparison.

• Decomposed a monolith into NestJS microservices and Serverless functions, creating infrastructure that supported future model-serving endpoints and scalable inference.
• Designed event-driven pipelines with AWS EventBridge and SQS to stream features and labels, enabling reliable data ingestion for ML dataset preparation.
• Optimized backend data pipelines and PostgreSQL queries to support low-latency retrieval, improving throughput for downstream model inference workloads.
• Implemented GitHub Actions CI/CD, Docker and Kubernetes on EKS to standardize deployments and create reproducible environments for model packaging and serving.
• Integrated Redis caching and Elasticsearch to speed retrieval for semantic search prototypes aligned with RAG experiments.

• Built a Node.js backend with PostgreSQL and RabbitMQ for background processing, enabling reliable asynchronous jobs for model preprocessing and batch inference.
• Implemented CI/CD and code quality tooling with GitLab, improving deployment reliability for services that feed ML pipelines.
• Delivered end-to-end features with React and Ruby on Rails that integrated with backend services used to collect labeled data and user feedback for model improvement.

• Integrated Elasticsearch and Redis caching to improve search latency, supporting semantic retrieval experiments for downstream RAG workflows.
• Built a GraphQL platform on Node.js that handled high-volume product data, providing structured inputs for feature extraction and dataset creation.
• Implemented scalable data pipelines and storage on MSSQL to support large ingestion jobs used in dataset curation for ML experiments.

A platform-agnostic cloud infrastructure components library for modern applications. Built with TypeScript and Pulumi, Infra Foundry provides reusable, composable infrastructure components that work across AWS, Cloudflare, and Vercel.

Built and engineered an end-to-end, 10-stage GTM engine for an AI-evaluation startup using TypeScript and GitHub Actions. Designed as an idempotent, Notion-backed state machine, the pipeline processes YC prospects via a provider-agnostic Claude client (Anthropic and AWS Bedrock via secure OIDC) with automatic model fallback and an automated, fail-closed IMAP reply-sweeper. To drive growth, I transformed pipeline artifacts into AI-generated evaluation rubrics that double as cold-outreach lead magnets, turning backend infrastructure into a high-leverage product-led growth loop.