John Carl Santos

An AI startup. Curating training datasets, defining annotation standards, and building evaluation sets for LLM and machine learning systems. Research the approach for upcoming client projects before a build starts, then build those systems end to end — LLM-backed document and email automation, and internal ticketing tooling — under NDA, so the specifics stay off this page.

Researched costs, model licences, and language coverage for a multilingual AI phone caller in Singapore. Estimated a Pipecat, Twilio, and Claude pipeline at $0.08–0.11 per minute against $0.14–0.18 for the alternatives evaluated. Prototyped member-scoped agent tools with capped refunds and checks for data leaks.

Built a RAG pipeline with vector search and LLM APIs, added JWT authorization, and containerized a multi-service backend.

Developed RESTful APIs, delivered backend features, and optimized application performance in an agile environment.

Building and evaluating deep learning models for classification and regression with PyTorch.

Consistent Dean's Lister with a 1.15 GWA. Coursework across software engineering, algorithms, linear algebra, and AWS cloud computing.