Filipino Devs Are Building Government AI Tools in Discord Because DICT Won't
While agencies wait for budget cycles and foreign vendors, small groups of Filipino developers are shipping Tagalog AI tools from group chats and weekend sprints.
Somewhere in a Discord server with maybe 200 active members, a Filipino developer is fine-tuning a small language model to understand Bisaya, Hiligaynon, and Tagalog code-switching. Nobody is paying him. He has a day job at a fintech startup. The model will probably end up free on Hugging Face by the end of the year.
This is how Filipino-language AI tools are getting built in 2026. Not by DICT. Not by DOST. Not by any agency with a procurement budget that runs into nine figures. By people in group chats, on weekends, between Slack messages from their actual jobs.
The procurement gap is a generation-defining problem
Government agencies that need AI tools, court transcription, barangay record digitization, public health chatbots, are mostly buying from foreign vendors or waiting for budget cycles that move slower than the typhoons. The tools that arrive are trained on English-heavy datasets and choke on actual Filipino input. Ask a generic chatbot to summarize a barangay blotter written in Taglish and watch it hallucinate names.
Filipino developers know this because they have tried to use these tools at work. So a few of them started building alternatives. Open-source Tagalog tokenizers. Small models trained on scraped Senate hearing transcripts. Speech-to-text experiments that actually recognize the way people talk in Quezon City, not the way a Berkeley grad student thinks they do.
The Discord servers are doing what agencies were budgeted to do
The setup is familiar to anyone who has watched Filipino devs build anything online. A pinned message with the project goals. A channel for dataset cleaning. Another for model evaluation. Volunteers labeling data on their lunch breaks. Someone in Cebu running training jobs on a rented GPU because nobody can afford a permanent one.
None of these projects have procurement contracts. None of them have signed MOAs with government. A few have been quietly used by NGOs and a couple of LGUs that figured out the open-source tools work better than what they were sold.
Meanwhile, agency websites still publish bid notices for AI projects priced in the tens of millions, awarded to consultancies that subcontract the actual work. Sometimes the subcontractors are the same people in the Discord server, getting paid a fraction of what the prime contractor billed.
What the volunteers are actually subsidizing
The romantic version of this story is that Filipino devs are scrappy and resourceful and that bayanihan lives on in GitHub repos. The less romantic version is that public infrastructure is being built for free by workers who already have jobs, because the institutions tasked with building it have not figured out how to procure software written this decade.
The volunteers are not waiting to get noticed. Most of them have stopped expecting government partnerships. The realistic ceiling is that a foreign company eventually hires them, or a local startup acquires the project, or the repo goes stale when the lead maintainer burns out.
The tools will exist. Filipino-language AI will get built. The question is who owns it when it does, and whether the developer who spent six months labeling Bisaya sentences after work ever sees a peso for it. The answer right now is no. The answer next year is probably also no. The Discord server keeps the lights on anyway.
