Darnel Castor
darnelcastor2@gmail.com | Github | Linkedin
Professional Summary
I'm an AI Engineer with specific expertise in LLM systems, reinforcement learning workflows, and production-grade backend architectures. My work involves optimizing AI-generated code, ensuring reliability in AI model outputs, and developing practical tools that businesses can use, not just demo. I'm currently refining AI-generated code by analyzing it using reinforcement learning workflows. This improves accuracy in AI model outputs, ensuring they align with real-world engineering best practices. My Computer Science background, combined with my experience in AI model training, full-stack development, and chatbot engineering, enables me to bridge the gap between experimentation and production.
Skills
- Languages: JavaScript, TypeScript, Python
- AI/ML: Tensorflow, Amazon BedRock, LangGraph, LangSmith
- Cloud & DevOps: Amazon Web Services
- Databases: PostgreSQL, Neo4j
- Frontend: Next.js
Experience
- Evaluated AI-generated code across Python, JavaScript, TypeScript, and Go, identifying runtime inefficiencies, logic bugs, and security vulnerabilities to cut errors by 30% and accelerate delivery of high-performing code-generation models.
- Engineered adversarial prompts and algorithmic test cases covering recursion, data structures, and dynamic programming, then benchmarked model outputs against human-verified reference solutions and unit tests to enforce strict performance and execution standards
- Reviewed, refactored, and annotated complex codebases, documenting precise technical feedback on code failures to produce high-quality training data for Reinforcement Learning from Human Feedback (RLHF) pipelines and teach the model optimized architectural patterns
- Full-stack engineer on a cross-functional team of technical and non-technical contributors, building a multi-cloud platform for evaluating and comparing LLMs at scale.
- Designed and shipped a multi-cloud LLM evaluation platform (AWS + GCP) supporting side-by-side comparison across 25 models, built with TypeScript, React, and PostgreSQL.
- Software Engineer Intern on Outreach's Orchestration Services team building data pipeline infrastructure and optimizing legacy database services.
Projects / Portfolio
AI Policy Copilot is a production-ready RAG (Retrieval-Augmented Generation) application I designed and built end-to-end, enabling natural-language search over government and corporate policy documents such as the EU AI Act and NIST AI RMF.