Open-Source AI Code Camps · Barangay AI

Build your own AI. 100% free and open-sourced.

At the Barangay AI Beta Code Camp on August 27, 2026, 19 builders went from “what is an LLM” to a live, public AI in one afternoon, on ordinary computers, with zero cloud spend required.

See the AI showcase Explore the platform

Beta results

Everyone finished. Everyone would come back.

Part of a nationwide push: 500+ builders trained on AI Fluency, and growing.

100%Completion: all 19 builders finished all 19 verified steps
100%Would tell a friend to join the next one
89%Plan to keep building on their AI
19Builders, each shipping a solo build

How the camp works

Four sprints. Nineteen verified steps.

A step-by-step checklist walks every builder from installing Ollama to shipping a live AI. Every step is time-boxed, auto-verified, and carries proof of work and facilitator notes, with no separate forms.

Understand and fork
Learn how LLMs work, then fork the Barangay AI repo.
Run it locally
Install Ollama and pull an open-weight model on your own machine.
Customize
Give your AI a name, a persona, a reply language, and your own documents.
Publish
Connect it to a real chat app and ship a live public URL.

Open the Code Camp Checklist

Screenshot of the Code Camp Checklist: step to install Ollama and pull a model — The Code Camp Checklist: every step verified and stamped in real time.

Build with open source

Ollama, GitHub, and Vercel. Free, no cloud needed.

Future-ready developers know how to pick the right tools and build with open source. Every builder forked the Barangay AI app and ran it on their own machine.

Local-first AI chat
Talks to a model running on your own machine through Ollama. Your chats never leave your device.
Speaks your language
Replies in English, Filipino, Taglish, Bisaya, Hiligaynon, or Ilocano.
Shows its receipts
Grounded on your own files, with the exact source behind every answer.
Works offline
Once loaded, the only thing it needs is your model.
See the code on GitHub (opens in a new tab)

Tools across our AI camps

OpenCode
Anthropic Claude
Ollama
Qwen
Gemma
Llama
GitHub
Vercel
and more

What they built

Personal, grounded in real documents, and live

Personas

6 companions like Liv AI, TALI, SKAD, and Gon
5 study tutors like Master Shifu, Auren, and Bagbag AI
4 personal builds made to tinker with
2 barangay assistants, including Pare Santiago for Brgy. Santiago
2 culture builds, ChizMakers and JaysonTheGrasscutter

Grounded in real files

19 of 19 brought their own document, nobody tested on a sample
Coursework: Circuit Theory, UI/UX prep, ITEC-85
Live projects: a CocoFiber capstone and Operation Longshot
Civic paperwork: a barangay consent form, queried in Tagalog

Models and choices

Open-weight models, served locally with Ollama

The top two models powered 84% of all builds. 16 of 19 builders picked the 1B to 3B size band; 3 sized up to 7B–8B.

Models chosen by builders

qwen2.5:3b14 builders · 73.7%
llama3.1:8b2 builders · 10.5%
qwen2.5:1.5b1 builder · 5.3%
gemma3:4b1 builder · 5.3%
qwen2.5:0.5b1 builder · 5.3%

Reply language configured

One in four replied in Filipino or Taglish.

English14
Taglish3
Filipino2

Cloud providers added

Optional, by the 5 builders who added a cloud key.

DeepSeek5
Groq4
OpenAI1

Every OS finished with no drop-off: 16 of 19 built on Windows, alongside macOS and Linux.

Showcased on stage

Three builders demoed their AI to the room

Builder
Jayson Ramos Hermogenes
AI name
JaysonTheGrasscutter
Try JaysonTheGrasscutter (opens in a new tab)
Builder
Julius Fuentiblanca
AI name
Gon
Try Gon (opens in a new tab)
Builder
Benedict Fusin
AI name
Auren
Try Auren (opens in a new tab)

A builder presenting their AI on stage in front of the code camp audience

Retrospective

What worked, and where we’ll focus next

What worked

100% completion across all 19 verified steps
100% positive sentiment: every builder would tell a friend
84% said the pace felt about right
79% of those who got stuck were unstuck within minutes
89% plan to keep building after the camp
Windows, macOS, and Linux all finished with no drop-off

Where to improve

Sprint 3 (Customize) was the hardest step for 42% of builders
Pre-install and local setup caused 37% of early friction
Pace was polarizing for a few: 2 said too fast, 1 too slow
More roaming facilitators, so no one falls behind
Clearer prompts so app versions aren’t mistaken for model names

Project directory

Every builder’s live demo, in one place

Builder	AI name	Live link
John Gabriel R. Batac	Master Shifu	Open live demo of Master Shifu (opens in a new tab)
John Ray Cacananta	Sinag	Open live demo of Sinag (opens in a new tab)
Benedict Fusin	Auren	Open live demo of Auren (opens in a new tab)
Aaron John Bitara	Kenyo	Open live demo of Kenyo (opens in a new tab)
Gian Carlo Gonzalvo	Rvngrr	Open live demo of Rvngrr (opens in a new tab)
James Fontanilla	Bagbag AI	Open live demo of Bagbag AI (opens in a new tab)
Shelley Sesante	Olla	Open live demo of Olla (opens in a new tab)
Bernadette Tormo	DEVCON	Open live demo of DEVCON (opens in a new tab)
Jasmine Mikaella Aninion	Liv AI	Open live demo of Liv AI (opens in a new tab)
Reymie Gonzaga	TALI	Open live demo of TALI (opens in a new tab)
James Andre L. Kalaw	Jalk AI	Open live demo of Jalk AI (opens in a new tab)
Kim Jerald S. Sialana	Pare Santiago	Open live demo of Pare Santiago (opens in a new tab)
Jhadz Ignacio	SKAD	Open live demo of SKAD (opens in a new tab)
Jayson Ramos Hermogenes	JaysonTheGrasscutter	Open live demo of JaysonTheGrasscutter (opens in a new tab)
Aje Balahula	beks	Open live demo of beks (opens in a new tab)
Julius Fuentiblanca	Gon	Open live demo of Gon (opens in a new tab)
Kate Anne Marie Dalangin	ChizMakers	Open live demo of ChizMakers (opens in a new tab)
Rhona Angelica Cailao	snooki	Open live demo of snooki (opens in a new tab)
Monnic Chelica Miles Ramos	Asul	Open live demo of Asul (opens in a new tab)