About
I'm Bart Porcincula. Senior AI Engineer.
I'm based in the Philippines. I build and evaluate LLM-based agents for production software, with TypeScript, Python and Go on the backend. I measure before I claim anything. I use a model only where the output is language or a judgement call, and keep every other decision deterministic. When the data shows one of my design records was wrong, I correct it.
Outside work I run local LLMs on a single consumer 16 GB GPU (llama.cpp, Ollama, GGUF models) and use it for small, pre-registered experiments on open models: LoRA fine-tuning, calibration, quantisation and serving. I publish the negative results as results. The research section has the numbers behind each write-up, and the projects section has longer technical write-ups.
Inside the lab
The current setup is an RTX 5060 Ti with 16 GB VRAM, a Ryzen 5 7500F and 32 GB RAM. The lab page records the hardware, software and limits behind the experiments.
See the rig and setup historyExperience
Senior AI Engineer
June 2026 – present
Software Developer (from August 2021), then Software Engineer (from January 2023), then Senior Software Engineer (from July 2023)
August 2021 – June 2026
Employer names are left off this site, and client work from these roles is not described here.
Education
BS Computer Engineering, Technological Institute of the Philippines, 2021.
Skills the work shows
- ▲LLM agents and evaluation: tool-calling loops, reference sets, control arms and noise floors.
- ▲Fine-tuning and serving: LoRA and QLoRA, ModernBERT classifiers, calibration (ECE, temperature scaling), GGUF quantisation, llama.cpp, Ollama, vLLM.
- ▲Experiment design: pre-registration, paired bootstrap confidence intervals, power checks, held-out leakage tests.
- ▲Backend and data: TypeScript, Next.js, Python, Go, PostgreSQL, Supabase, DuckDB, dbt, Dagster.
- ▲Privacy: PII de-identification, GDPR and UK GDPR, audit-trail design.