← Resources

// AI Week HK 2026

Build with Votee.

An open-source, production-grade stack for your hackathon — plus a morning workshop, on-site mentors, and a Votee prize.

5 July 2026 (Sun)) · Hong Thai Expo, 5/F, 95 Queensway, Admiralty, Hong Kong

Votee workshop — morning, at the start of hacking

We’ll walk you through the tools, get you running in minutes, and show the fastest path to a working build.

// Get Running

Running in 60 Seconds

Option A · pre-seeded demo

No API keys — best for a first look.

git clone https://github.com/Beever-AI/beever-atlas.git
cd beever-atlas
make demo

Fully seeded instance (Q&A, entity graph, wiki) in under a minute — zero keys.

Option B · full guided install

Bring your own data.

git clone https://github.com/Beever-AI/beever-atlas.git
cd beever-atlas
./atlas

Guided setup (pick providers, paste a free key), opens at http://localhost:3000.

What to bring

Docker + Docker Compose. A free GOOGLE_API_KEY(Gemini — no credit card, 2-min signup) is only needed for live Q&A / wiki generation, not for make demo.

// The Build Target

Beever Atlas

Open-source memory layer for AI agents · Apache 2.0

Turns scattered team chat across Slack, Microsoft Teams, Discord, Telegram and Mattermost into a living memory layer for AI agents — searchable, retrievable, and usable by any assistant via the Model Context Protocol (MCP). Open-source, self-hostable, plugs straight into agents.

GitHub · beever-atlas →Product · beever.ai →

Also available — optional, for data-focused teams

MAGIC data-agent skills

Open-source data-engineering skills (parsing, extraction, transformation) you can compose into pipelines — useful if your build processes messy real-world data.

github.com/Votee-AI/magic-agent-skills →docs.votee.ai/magic-agent-skills →

Lingua Corpora

An interactive live map of how much data exists across the world’s languages — a reference to pick a neglected language to champion and find where its data lives.

lingua.beever.ai →

// The Ask

What We Hope You'll Build

  • ◆Something real and working — a demo that runs beats a deck.
  • ◆Lean into sovereignty — apps where owning the model/data matters (on-prem, private, regulated).
  • ◆Use the memory layer / MCP in a non-obvious way — agents that get smarter from history.
  • ◆Champion neglected languages — especially Cantonese (see the honest note below).
  • ◆Bonus: open-source your own hackathon project too.

// Starting Points

8 Use Cases — All Doable in 1–2 Days

Each has a starter spec in the builder guide.

Build on Beever Atlas (memory + MCP) — recommended for most teams

  1. 1

    A team-onboarding agent answering new joiners' questions from Slack/Discord history.

  2. 2

    A “why did we decide X?” tracker that retrieves the reasoning behind past decisions.

  3. 3

    A community-support bot for a Discord/Telegram group that remembers prior threads.

  4. 4

    Plug Atlas into a coding assistant via MCP so it recalls your team's past choices.

  5. 5

    Meeting-memory — turn standups into retrievable action items.

Data track (MAGIC data skills + open language data)

  1. 6

    A data-processing pipeline that cleans/structures messy multi-format data into something an agent can use.

  2. 7

    A dataset-builder for a low-resource language — collect, clean and structure Cantonese (or another underserved language) data.

  3. 8

    Data-storytelling that makes the “AI Data Desert” tangible — use Lingua Corpora to choose a language and source the data.

Honest note — what we provide vs. what you bring

Beever Atlas is self-hostable and runs in minutes (make demo, no keys). The Cantonese model is open-weights on Hugging Face to download — there’s no hosted inference endpoint and we can’t provide GPUs, so a live Cantonese-LLM build means bring your own hosting. MAGIC data-skills is a code repo you run locally. When in doubt, build on Atlas — the smoothest path to a working demo.

// Prizes & Help

What You Can Win

🏆 Votee Track Prize

Best Use of Votee’s Open-Source Sovereign AI Stack · value TBC

Judged on: a working build · meaningful use of sovereignty/on-prem or a neglected language · authentic engagement with the open-source project.

General tracks — organiser to confirm.

Get help

  • 🗓️ Morning workshop at the start of hacking — tools, setup, fast path to shipping.
  • 📖 Builder starter guide — a running start for all 8 use cases.
  • 🧑‍💻 Votee / Beever engineers on-site all day.

Why this matters

Votee’s mission is Sovereign AI for the world’s neglected languages — accurate AI for the 3 billion people underserved by today’s global models. These tools are how we put that into practice. Build with us. 🇭🇰

votee.ai →beever.ai →