Votee AI

[Votee AI @Votee_AI]

Votee AI is on @BBC! 🎙️Global AI has a "blind spot" for languages like Cantonese. Our CEO @paksunting sat down with @BBCWorld Tech Life to share how we're building "Gov-Enterprise Grade" Sovereign AI to serve 80M+ speakers & beyond. 🎧Starts at 18:55:
https://t.co/Akg3JPnjAl

[Votee AI @Votee_AI]

𝗧𝗵𝗮𝘁'𝘀 𝗮 𝘄𝗿𝗮𝗽 𝗼𝗻 𝗧𝗼𝗿𝗼𝗻𝘁𝗼 𝗧𝗲𝗰𝗵 𝗪𝗲𝗲𝗸 𝟮𝟬𝟮𝟲. 🦫

The Beever AI Toronto Showcase brought banks, insurers, government, VCs, academia, and media into one room — We were thrilled to see a full house with every seat occupied. 👥👥

This success was made possible through the generous support of our sponsors: 𝗕𝗗𝗢 𝗖𝗮𝗻𝗮𝗱𝗮, 𝗔𝗹𝗶𝗯𝗮𝗯𝗮 𝗖𝗹𝗼𝘂𝗱, 𝗤𝘄𝗲𝗻, and 𝗧𝗼𝗿𝗼𝗻𝘁𝗼 𝗧𝗲𝗰𝗵 𝗪𝗲𝗲𝗸. We would also like to thank all of our esteemed attendees for their participation and engagement. ❤️🙏

What we put on stage:
🦫 𝗕𝗲𝗲𝘃𝗲𝗿 𝗔𝘁𝗹𝗮𝘀 — our open-source (Apache 2.0) memory layer that turns team chat into searchable, living knowledge.
🔗https://t.co/RLpzxttiMg
🔗https://t.co/0v2y1rfYo9

🧠 𝗩𝗼𝘁𝗲𝗲 𝗠𝗔𝗚𝗜𝗖 — enterprise LLM training you fully own, on-prem. (The engine behind the world's first fully pre-trained Cantonese LLM.)
🔗https://t.co/RpSqzjfqRH
🔗https://t.co/vPanTEM7T1

🐎 𝗛𝗮𝗽𝗽𝘆𝗛𝗼𝗿𝘀𝗲 — live AI video generation on Alibaba Cloud Model Studio, where every attendee made their own.
🔗https://t.co/EswpQA0TBM

The conversations are just getting started.

Contact our team today to secure your AI roadmap:
📩 [email protected]

Built by Toronto, for the world.

[Votee AI @Votee_AI]

𝗧𝗼𝗺𝗼𝗿𝗿𝗼𝘄 𝗶𝘀 ��𝘂𝗿 𝗹𝗮𝘀𝘁 𝗱𝗮𝘆 at AI+ Power 2026 (Presented by BGOV) — if you’re at HKCEC, come by and meet us. 🤖🚀

We’ve been connecting with teams building real AI for the business world—and we want to keep the momentum going through tomorrow.

What we put on the floor (same products as our demo stack):
🦫 𝗕𝗲𝗲𝘃𝗲𝗿 𝗔𝘁𝗹𝗮𝘀 — our open-source (Apache 2.0) memory layer that turns team chat into searchable, living knowledge.
🔗https://t.co/RLpzxttiMg
🔗https://t.co/0v2y1rfYo9

🧠 𝗩𝗼𝘁𝗲𝗲 𝗠𝗔𝗚𝗜𝗖 — enterprise LLM training you fully own, on-prem. (The engine behind the world's first fully pre-trained Cantonese LLM.)
🔗https://t.co/RpSqzjfqRH
🔗https://t.co/vPanTEM7T1

🐎 𝗛𝗮𝗽𝗽𝘆𝗛𝗼𝗿𝘀𝗲 — live AI video generation on Alibaba Cloud Model Studio, where every attendee made their own.
🔗https://t.co/EswpQA0TBM

Contact our team to build your AI roadmap:
📩 [email protected]

[Votee AI @Votee_AI]

Every fund has bought into AI. The ones turning it into alpha are rebuilding their data workflows from the language up.

Frontier models claim to read 200 languages. The ones APAC desks actually trade on — Cantonese earnings calls, Korean Naver threads, Japanese Yahoo boards — still get mistranslated at the level that costs you the signal.

🎙️ Our Chief Scientist APAC, Dr. Pui-Wai (Leo) Ma, joined the Data Workflows panel at the Hong Kong Data Summit 2026 by @neudatalab (Grand Hyatt, Wan Chai) — alongside Min Fu (Head of Asia Data Research, @p72vc) and Kevin Sung (Head of ETF Portfolio Management, @miraeasset), moderated by Nathaniel Rushforth (Senior Counsel, Shan Zhang International Law Firm).

Leo's framing:
"Most teams treat the language layer as a translation problem. It's a context problem. Translation gets you the words; context gets you the trade."
A model can summarise an earnings call into clean English and lose the Cantonese hedge a local analyst hears in five seconds. That gap is the trade.

We're building the layer that gets it right — Beever Atlas: sovereign AI with memory, in your own jurisdiction, that keeps a desk's institutional reasoning when the analyst who heard it leaves.

More on what we're building for APAC capital markets ↓
🔗 https://t.co/J1zidaWSKD

#SovereignAI #Cantonese #AlternativeData

[Votee AI @Votee_AI]

Beyond the chatbot 🎙️

Our Head of Sales Jeff Tai took the stage at
@cyberport_hk 's AI Frontier Forum 🇭🇰

Every enterprise has bought into AI.
Most are still waiting for it to pay off ⏳

The model isn't the problem.
It's the context, the memory, and the human
judgement around it.

🧱 We're building those layers.

—

3 takeaways from the stage 👇

1️⃣ Teach AI your context
Generic models trip on Cantonese homophones —
"燕梳" means insurance, "收口" means finishing.
That last 5% of tuning is the whole game.

2️⃣ Adoption is a journey
Yesterday's frontier is tomorrow's commodity.
You can't train AI once and use it forever ⚡️

3️⃣ Memory is the missing layer
Beever Atlas turns Slack, Teams, Gmail and
meeting recordings into a queryable knowledge
graph — plugged into Claude, Cursor and MCP 🧠

—

The line that stuck with the room 🔥
"Empathy, agents, and memory — the three pillars
of AI that actually works."

—

🙏 Thank you @cyberport_hk for the stage.
📩 Enterprise AI roadmap?
→ https://t.co/J1zidaWSKD

[Votee AI @Votee_AI]

99.9% of the world's 7,000+ spoken languages are still locked out of the AI revolution. The major foundation models all converge on the same handful of languages. The rest of humanity gets translation, not understanding.

We're building for the rest.

Our co-founder Pak-Sun Ting sat down with the Toronto Star to talk about what we're building from our R&D hive in Toronto and our HQ in Hong Kong:

→ The first Hong Kong Cantonese LLM — for 8 million native speakers whose language code-switches with English in ways that break most models
→ Beever Atlas: our open-source platform that turns conversations into a structured, self-maintaining wiki — unveiled this week at Toronto Tech Week
→ Extending the work to First Nations, South Asian, and East African languages
→ A research collaboration on therapy-training avatars for clinicians working with emo­tional issues

On translation alone not being enough, Pak put it plainly:
"You need to not just focus on the data but also the code level. Go deep into that lan­guage."

On AI's broader impact — the very real jobapocalypse anxieties this column doesn't dodge — Pak didn't dodge either:
"It's the global nature of geo­pol­it­ics and also the trans­ition from cer­tain jobs being taken away ... I do think job loss and this whole trans­ition of the eco­nomy is a true thing. But if you delay this trans­ition, the job loss may be even worse in the future. This is a mech­an­ism of how the eco­nomy works, to reflect the changes of soci­ety, change in job­makers."

And on whether any of this is stoppable:
"It's hard to stop innovation. It's very hard to stop intelligence."

Grateful to the Toronto Star for the thoughtful reporting on what Sovereign AI for neglected languages actually means — beyond the hype, beyond the headlines, all the way down to dim sum that AI still can't translate as "touch the heart."

Read the full piece: https://t.co/T5R08QuRCP

#SovereignAI #LLM #DeepTech #Cantonese #TorontoTechWeek #NeglectedLanguages #LowResourceLanguages

[Beever AI @Beever_AI]

99.9% of the world's 7,000+ spoken languages are still locked out of the AI revolution. The major foundation models all converge on the same handful of languages. The rest of humanity gets translation, not understanding.

We're building for the rest.

Our co-founder Pak-Sun Ting sat down with the Toronto Star to talk about what we're building from our R&D hive in Toronto and our HQ in Hong Kong:
→ The first Hong Kong Cantonese LLM — for 8 million native speakers whose language code-switches with English in ways that break most models
→ Beever Atlas: our open-source platform that turns conversations into a structured, self-maintaining wiki — unveiled this week at Toronto Tech Week
→ Extending the work to First Nations, South Asian, and East African languages
→ A research collaboration on therapy-training avatars for clinicians working with emo­tional issues

On translation alone not being enough, Pak put it plainly:
"You need to not just focus on the data but also the code level. Go deep into that lan­guage."

On AI's broader impact — the very real jobapocalypse anxieties this column doesn't dodge — Pak didn't dodge either:
"It's the global nature of geo­pol­it­ics and also the trans­ition from cer­tain jobs being taken away ... I do think job loss and this whole trans­ition of the eco­nomy is a true thing. But if you delay this trans­ition, the job loss may be even worse in the future. This is a mech­an­ism of how the eco­nomy works, to reflect the changes of soci­ety, change in job­makers."

And on whether any of this is stoppable:
"It's hard to stop innovation. It's very hard to stop intelligence."

Grateful to the Toronto Star for the thoughtful reporting on what Sovereign AI for neglected languages actually means — beyond the hype, beyond the headlines, all the way down to dim sum that AI still can't translate as "touch the heart."

Read the full piece: https://t.co/aUNxsKvYf1

@TorontoStar

#SovereignAI #LLM #DeepTech #Cantonese #TorontoTechWeek

[Beever AI @Beever_AI]

99.9% of the world's 7,000+ spoken languages are still locked out of the AI revolution. The major foundation models all converge on the same handful of languages. The rest of humanity gets translation, not understanding.

We're building for the rest.

Our co-founder Pak-Sun Ting sat down with the Toronto Star to talk about what we're building from our R&D hive in Toronto and our HQ in Hong Kong:
→ The first Hong Kong Cantonese LLM — for 8 million native speakers whose language code-switches with English in ways that break most models
→ Beever Atlas: our open-source platform that turns conversations into a structured, self-maintaining wiki — unveiled this week at Toronto Tech Week
→ Extending the work to First Nations, South Asian, and East African languages
→ A research collaboration on therapy-training avatars for clinicians working with emo­tional issues

On translation alone not being enough, Pak put it plainly:
"You need to not just focus on the data but also the code level. Go deep into that lan­guage."

On AI's broader impact — the very real jobapocalypse anxieties this column doesn't dodge — Pak didn't dodge either:
"It's the global nature of geo­pol­it­ics and also the trans­ition from cer­tain jobs being taken away ... I do think job loss and this whole trans­ition of the eco­nomy is a true thing. But if you delay this trans­ition, the job loss may be even worse in the future. This is a mech­an­ism of how the eco­nomy works, to reflect the changes of soci­ety, change in job­makers."

And on whether any of this is stoppable:
"It's hard to stop innovation. It's very hard to stop intelligence."

Grateful to the Toronto Star for the thoughtful reporting on what Sovereign AI for neglected languages actually means — beyond the hype, beyond the headlines, all the way down to dim sum that AI still can't translate as "touch the heart."

Read the full piece: https://t.co/aUNxsKvYf1

@TorontoStar

#SovereignAI #LLM #DeepTech #Cantonese #TorontoTechWeek

[Votee AI @Votee_AI]

Votee AI in @FortuneMagazine.

Cofounder Pak-Sun Ting on why accuracy in low-resource languages isn't a niche — it's the precondition for AI to work in regulated industries.

https://t.co/L5B7JSjTnw

[Votee AI @Votee_AI]

Ready to scale your AI securely? Let's talk → [email protected]

Beever Atlas — turns your Telegram, Discord, Mattermost, MS Teams & Slack chats into a living wiki:
https://t.co/QgrmiPkjkU

[Votee AI @Votee_AI]

"Generic models hit a wall when they go to nuanced spoken language."

— @paksunting, CEO @Votee_AI, on @TheAIInnov

Why Big Tech won't build precise AI for the billions underserved by current models — and why we will:

https://t.co/cLd9JSOVSB

[Votee AI @Votee_AI]

80M Cantonese speakers.
~2,000 Southeast Asian languages.
~1,000 African languages.

Banks need 85-95% accuracy. Generic global LLMs don't get there.

[Beever AI @Beever_AI]

"Mom sent me a recipe. Telegram."
Three months ago. I can't find it.

A friend recommended a Tokyo sushi spot.
I'm flying there next week.
Where's that message?

"Why did we pick Stripe over Adyen?"
Buried in Slack.

Every conversation you've ever had —
the answers are in there.

But we can't find them.

Beever Atlas turns your chat history into an organized wiki — automatically.

Free. Open source. Runs on your laptop.

⭐ https://t.co/61A5VbpXln

For Enterprise Version: [email protected]

[Beever AI @Beever_AI]

The agent has 10 MCP tools.
And ZERO router code.

How does it decide what to do?

Three answer modes — same brain, three surfaces:

→ Quick (5s): 2 wiki tools, planner disabled, single citation
→ Summarize (10s): 4 tools + search history, two-paragraph synthesis
→ Deep (30s): all 10 tools, full orchestration, the complete picture

Same question. Different mode. Different cost, depth, answer.

The dispatcher instinct says: classify the query, route to the right handler.

We resisted that instinct.

The architecture: one LlmAgent, a list of tools, a system prompt.
The routing IS the prompt.
Tool descriptions do the heavy lifting.

Drop a new tool with a good docstring → the agent uses it immediately.
No routing logic to update.

The article's punchline:
"Resist the dispatcher instinct until tool descriptions stop working.
They'll take you further than you expect."

⭐ https://t.co/61A5VbqvaV

Apache 2.0 · Open source · runs on your laptop.

[Votee AI @Votee_AI]

Moving from AI Innovation to Industrial-Grade Retail Practice

Retail is no longer just about the storefront; it's about the Sovereignty of Data — and the security of the agents acting on it.

At the 2026 Retail Innovation Forum, our Head of Sales, Jeff Tai, joined industry leaders, Tony Pow of HansTech, Kinman T. of Maxim's Caterers Ltd and Martin Ip of @CheckPointSW to deconstruct a critical challenge: How can retail enterprises bridge the gap between AI experimentation and secure, large-scale implementation?

The consensus was clear: Generic AI is a risk retail giants can no longer afford to take. As Hong Kong advances its new cybersecurity insurance frameworks and critical infrastructure protections, the "Blind Spot" of data privacy — and the emerging attack surface of AI agents themselves — has become a trillion-dollar frontier.

Key takeaways from the panel:

🔒 Sovereignty is the Moat: In a "Data Desert," retail leaders are turning to Sovereign AI. By deploying AI 100% on-premises, enterprises ensure that sensitive customer insights remain an internal asset, not a public liability.

🛡️ AI Security is the New Perimeter: The threat model has changed. Prompt injection, model poisoning, agent hijacking, and shadow AI tool-calls are now board-level risks. Enterprise Grade deployment means Zero Data Leakage by architecture — every agent action audited, every inference traceable. AI security can't be bolted on; it has to be engineered in from the first line of code.

⚙️ Labor Transformation: AI isn't just optimizing processes; it's reshaping the labor structure. By automating high-volume customer inquiries and document processing, we allow human talent to focus on high-value strategy and "Vibe Coding" innovation.

Whether it's navigating the complexities of Cantonese slang in customer service or ensuring compliance in the Greater Bay Area, Votee AI is proud to be the "Government-Enterprise Grade" partner for the next wave of global intelligence.

Ready to scale your AI securely? Let's talk.

📩 [email protected]
🌐 https://t.co/85u7rObCox

#AgenticAI #SovereignAI #EnterpriseAI #DigitalTransformation

#HongKong #CIO #AIStrategy #VoteeAI #CantoneseAI

[Votee AI @Votee_AI]

Two weeks ago @karpathy posted about "LLM Knowledge Bases" — the idea that LLMs should maintain structured, evolving knowledge from your documents. 1.7M+ views. He said:

"I think there is room here for an incredible new product instead of a hacky collection of scripts."

We've been building exactly that. Today we're open-sourcing Beever Atlas. GitHub: https://t.co/sb2p3WPsNX

The difference: Karpathy's approach starts with manual file uploads. Beever Atlas starts with your team's chat. Slack, Discord, Teams, Telegram — the messy, unstructured conversations where 90% of organizational knowledge actually lives and dies.

Here's what it does:
- Connect your chat platform (self-service, takes 2 minutes)
- Ingestion pipeline extracts entities, facts, and relationships automatically
- Builds a Neo4j knowledge graph — not just text cross-references, actual typed relationships between people, projects, technologies, decisions
- Generates a living Wiki — DeepWiki-style, with topic hierarchies, concept maps, glossaries. Updates every sync.
- Ships as an MCP server — Claude, Cursor, any AI assistant can query your team's collective knowledge directly

From our internal deployment (4 Slack channels):
- 854 structured memories
- 1,899 entities
- 5,271 relationships
- 222 wiki entries auto-generated

What Karpathy built is single-user, requires Obsidian + CLI, text-only. Beever Atlas is multi-user, zero-install web UI, knowledge graph, MCP-native.

We built this at Beever AI, a Toronto-based research lab under Votee AI, because we needed it ourselves — our engineering team's context was scattered across Slack threads nobody reads. Now our agents can actually reason over what the team knows.

100% on-premise. Docker stack. Bring your own LLM via LiteLLM (Ollama, Gemma 4, whatever you run locally). Zero data leakage.

Turn your team's chat into a living wiki.

⭐ https://t.co/sb2p3WPsNX
💬 https://t.co/wBBtJq2wTv
🌐 https://t.co/kCBtv9Kun3

Shipped by the whole team:
Engineering — @jhkchan @cch_thomas @KaiYamYang1 @dantelok1111
Design — Adrian Leung
Comms & Media — @nghoihin

@Beever_AI is a Toronto-based research lab under @Votee_AI.

[Beever AI @Beever_AI]

Two weeks ago @karpathy posted about "LLM Knowledge Bases" — the idea that LLMs should maintain structured, evolving knowledge from your documents. 1.7M+ views. He said:

"I think there is room here for an incredible new product instead of a hacky collection of scripts."

We've been building exactly that. Today we're open-sourcing Beever Atlas. GitHub: https://t.co/61A5VbpXln

The difference: Karpathy's approach starts with manual file uploads. Beever Atlas starts with your team's chat. Slack, Discord, Teams, Telegram — the messy, unstructured conversations where 90% of organizational knowledge actually lives and dies.

Here's what it does:
- Connect your chat platform (self-service, takes 2 minutes)
- Ingestion pipeline extracts entities, facts, and relationships automatically
- Builds a Neo4j knowledge graph — not just text cross-references, actual typed relationships between people, projects, technologies, decisions
- Generates a living Wiki — DeepWiki-style, with topic hierarchies, concept maps, glossaries. Updates every sync.
- Ships as an MCP server — Claude, Cursor, any AI assistant can query your team's collective knowledge directly

From our internal deployment (4 Slack channels):
- 854 structured memories
- 1,899 entities
- 5,271 relationships
- 222 wiki entries auto-generated

What Karpathy built is single-user, requires Obsidian + CLI, text-only. Beever Atlas is multi-user, zero-install web UI, knowledge graph, MCP-native.

We built this at Beever AI, a Toronto-based research lab under Votee AI, because we needed it ourselves — our engineering team's context was scattered across Slack threads nobody reads. Now our agents can actually reason over what the team knows.

100% on-premise. Docker stack. Bring your own LLM via LiteLLM (Ollama, Gemma 4, whatever you run locally). Zero data leakage.

Turn your team's chat into a living wiki.

⭐ https://t.co/61A5VbpXln
💬 https://t.co/XzLTuPyWXy
🌐 https://t.co/pQf9RCfjXc

Shipped by the whole team:
Engineering — @jhkchan @cch_thomas @KaiYamYang1 @dantelok1111
Design — Adrian Leung
Comms & Media — @nghoihin

Beever AI is a Toronto-based research lab under @Votee_AI.

[Votee AI @Votee_AI]

Retraining a billion-parameter model from scratch every time you need new capabilities is not engineering. It's waste.

Today at The Hong Kong University of Science and Technology's CSE Research & Technology Fair 2026, our Senior AI Research Engineer Dante Lok presented "From Child to Adult: Growing Neural Networks Without Forgetting" — sharing Votee AI's original research into a fundamental challenge in large-scale AI deployment.

The core question: Can you expand a trained neural network — add neurons, add layers — without destroying what it already knows?

What Dante's research found:

Existing continual learning approaches often face a fundamental trade-off: either preserving prior knowledge at the cost of limited adaptability, or enabling adaptation while degrading previously learned functions.

He explores a different direction based on structured model growth, where the original function is preserved exactly at the moment of expansion, while new degrees of freedom are introduced to support subsequent learning.

This leads to a framework in which knowledge retention and acquisition are no longer treated as competing objectives, but as components of a unified process governed by the geometry of the parameter space. In particular, the interaction between growth and learning can be understood through how new dimensions are introduced and activated without interfering with existing representations.

Rather than relying purely on constraints or replay mechanisms, this perspective reframes continual learning as a problem of controlled evolution, where models expand and adapt in a principled manner.

While still early, this line of work suggests a more general view of neural networks as dynamically evolving systems, and would extend naturally across architectures such as MLPs, ResNets, Transformers and MoE.

Congratulations Dante and the Votee AI Research team.

Ready to scale your AI securely? Let's talk.

[email protected]
https://t.co/85u7rOcae5

#AgenticAI #HKUST #SovereignAI #EnterpriseAI #DigitalTransformation #HongKong #AIStrategy #VoteeAI #CantoneseAI

[Votee AI @Votee_AI]

Most enterprises say they're "doing AI." Most are stuck at Level 2.

Our Chief Scientist APAC Dr. Pui-Wai (Leo) Ma shared a strategic roadmap with CIOs and IT leaders from banking, insurance, and financial services at a closed-door CIO Collaboration Luncheon co-organised by TEH Group and Red Hat — and the core message was direct:

The shift from automation to autonomous intelligence is not a feature upgrade. It is an architectural transformation.

Dr. Ma outlined 7 Levels of AI Adaptation — from rule-based scripting to full Agentic Swarms — and the critical inflection point most organisations miss: the gap between Level 4 (builder tools) and Level 5 (reasoning engines). Below Level 5, systems generate text. Above Level 5, systems take action.

For enterprises operating in regulated environments — banking, government, insurance — this distinction is not academic. It determines whether AI delivers measurable ROI or becomes another stalled pilot.

Industry data shared during the session reinforced the urgency:
→ 171% projected ROI for Agentic AI (DigitalApplied, 2025)
→ 49% realised ROI already observed for Generative & Agentic AI (Snowflake / Omdia, 2026)
→ 15,000+ monthly hours saved across enterprise deployments (Lucidworks, 2025)

Dr. Ma also presented a practical four-phase deployment path — from Quick Wins in Week 1 through to full Agentic Transformation by Month 12 — demonstrating how our platform stack delivers compounding returns across research, data, compliance, and decision-making.

At Votee AI, we have deployed Sovereign AI solutions across enterprises — from DBS to the HKMA FSS 3.1 Pilot — because we believe Enterprise Grade AI must be 100% on-premise, zero data leakage, and built for the languages your workforce actually speaks.

The competitive advantage belongs to those who act now — not those who wait for the technology to "mature." It already has.

Thank you to the @TehGroupAsia @the_tehgroup team for hosting, and to @RedHat for co-organising a genuinely substantive conversation with Hong Kong's IT leadership community.

Ready to scale your AI securely? Let's talk.
[email protected]
https://t.co/85u7rObCox

#AgenticAI #SovereignAI #EnterpriseAI #DigitalTransformation #HongKong #CIO #AIStrategy #VoteeAI #CantoneseAI

[Votee AI @Votee_AI]

We just welcomed our youngest AI Security Research Apprentice. He's 16.

Jason Mak is a Form 4 student in Hong Kong. Before we invited him to join Votee AI, he had already found a way to jailbreak a language model — hiding adversarial commands in Hex code and using role-play exploits to bypass safety guardrails.

Then he went further.

Jason identified what we call the Linguistic Bypass — a critical security gap where global AI models fail to recognise the cultural weight of Hong Kong slang. An attacker can wrap harmful intent in expressions like "C9" or "零尊", and a global filter treats it as harmless. A native Cantonese model catches it immediately.

This is not a theoretical exercise. This is the frontline of Sovereign AI security.

Following his proactive disclosure, we officially appointed Jason as our AI Security Research Apprentice — contributing to Red Teaming stress-tests for CantoneseLLM-v2 alongside our enterprise security team.

Why this matters:
Hong Kong-built AI must be secured by Hong Kong talent. Not outsourced. Not translated. Not dependent on global models that are functionally illiterate in our own language.

We are building a sovereign security research pipeline — from secondary school to enterprise. Because the next generation of AI threats won't wait for the next generation to graduate.

Welcome to the team, Jason.

With Pak-Sun Ting @jhkchan Joyce Chan Pui-Wai (Leo) Ma Jeff Tai @nghoihin Katy C. Gariel Weng Dante Lok Adrian Leung Ching Lok Yu Ho Ching Sit Coco Ding Carson Ting

#SovereignAI #AIRedTeaming #CantoneseAI #VoteeAI #AISecurityResearch #HongKong #NeglectedLanguages #NextGenTalent