Mark Spencer J.

I build production AI systems that automate real business operations, from multilingual customer service pipelines to interactive voice agents, prompt security frameworks, and evaluation infrastructure. My work spans LLM agents, voice AI, RAG systems, workflow automation, and model evaluation. I’ve built and shipped 27+ production AI agents across 36 international markets, designed prompt injection security test suites with 100% pass rates, and engineered automated evaluation pipelines with 100-case benchmarks per agent. Recently, I turned my personal portfolio into an interactive voice experience where visitors can talk to the site instead of just scrolling through it. The voice agent can answer questions about my work, skills, experience, and availability, while also navigating the page in real time, highlighting projects, opening my Calendly booking flow, and using my actual portfolio context. Under the hood, I worked with Vapi, ElevenLabs, function calling, usage limits to protect voice credits, prompt guardrails, LLM evals, and prompt injection handling. It started as a portfolio, but became a working demo of the kind of AI systems I enjoy building: practical, interactive, and designed with production behavior in mind. Beyond LLM and voice systems, I’ve developed RAG pipelines using Pinecone, curated 30+ prompt templates for clinical applications, built a chatbot that reached 10,000+ visits, shipped an AI-powered Python IDE with sub-1-second completions, and optimized computer vision models to 98% classification accuracy under strict memory constraints. Core expertise: Prompt Engineering, LLM Pipeline Architecture, AI Agents, Voice AI, Vapi, ElevenLabs, Function Calling, NLP, RAG, Multi-Agent Systems, LangChain, Computer Vision, Python, Evaluation Frameworks Background: CS graduate from DLSU-Dasmariñas with a 3.7 GPA, Institutional Thesis Competition Overall Champion against 100+ theses, and Top Rated Upwork contractor. I treat every project like it’s my own product. Let’s talk if you’re building something that needs to work.