01
LLM Finetuning
Filipino-context adaptation of open-weight base models. Instruction tuning, LoRA, evaluation harnesses built on local corpora.
WHY
LOCAL WEIGHTS
Frontier models handle English fine and Tagalog badly. Bisaya, Cebuano and code-switched Taglish fall off a cliff. We build and finetune for the way Filipinos actually write.
11Languages in scope
PHContext focus
24/7Inference uptime target
OSSBase model policy
WORK
WHAT WE SHIP
02
Synthetic Data
Generated and human-audited datasets covering Cebuano, Tagalog and Taglish register, plus domain sets for news and tourism.
03
AI Post Pipelines
Transcription, translation, rough-cut assembly and metadata tagging folded into existing edit workflows.
04
Content Systems
Retrieval and summarisation layers feeding Balita.ph and internal newsroom tools.
05
Evaluation
Benchmarks that test for local idiom, not just BLEU. Human raters from the target region.
06
Advisory
Model selection, cost modelling and deployment guidance for PH-market teams.
LATEST
2024 → 2026
Two years of our GenAI work in the Philippines, remade with today's models. First up: a Visayan warrior who went from an AI drawing to a textured, rigged and animated 3D character in one browser tab.
READ THE WORK LOG →STACK
TOOLING
Open weights where possible, hosted inference where it makes sense, always with an eye on cost per token in a peso-denominated market.
More on what we do with AI, and the 3D worlds it builds: AI & 3D.
← PHILIPPINES