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: lnkd.in/gvpgeTFC

Ready to scale your AI securely? Let's talk → hello@votee.com Beever Atlas — turns your Telegram, Discord, Mattermost, MS Teams & Slack chats into a living wiki: github.com/Beever-AI/beev…

80M Cantonese speakers. ~2,000 Southeast Asian languages. ~1,000 African languages. Banks need 85-95% accuracy. Generic global LLMs don't get there.

"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: theaiinnovator.com/ais-language-g…

"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. ⭐ github.com/Beever-AI/beev… For Enterprise Version: hello@votee.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." ⭐ github.com/Beever-AI/beev… Apache 2.0 · Open source · runs on your laptop.

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. 📩 hello@votee.com 🌐 lnkd.in/gZz-6Hab #AgenticAI #SovereignAI #EnterpriseAI #DigitalTransformation #HongKong #CIO #AIStrategy #VoteeAI #CantoneseAI

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: github.com/Beever-AI/beev… 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. ⭐ github.com/Beever-AI/beev… 💬 discord.gg/VshBCUUX 🌐 beever.ai 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.

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. hello@votee.com lnkd.in/gZz-6Hab #AgenticAI #HKUST #SovereignAI #EnterpriseAI #DigitalTransformation #HongKong #AIStrategy #VoteeAI #CantoneseAI

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. hello@votee.com lnkd.in/gZz-6Hab #AgenticAI #SovereignAI #EnterpriseAI #DigitalTransformation #HongKong #CIO #AIStrategy #VoteeAI #CantoneseAI

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

Happy Easter! 🐰 | Bridging the AI Divide Dear friends and partners, Wishing you and your family a very happy Easter! May this season of renewal and new beginnings bring you joy, hope, and prosperity. As we celebrate this time, we look forward to achieving great things together. We're grateful for your trust and collaboration. At Votee AI, we are dedicated to bridging the global "Digital Divide" by building high-fidelity, native-language models for enterprises and regulated sectors—ensuring that the 3 billion people speaking neglected languages are not left behind by the AI revolution. Learn about our journey in the media: * BBC - A tech company developing AI for neglected languages (Starting at ): lnkd.in/g6jvx98P * Fortune - Why non-English languages risk falling further behind: lnkd.in/gXtkXxfs * South China Morning Post SCMP - How AI is Helping to Preserve Cantonese: Votee AI defines the scientific standard for regional languages and aims to replicate its blueprint for other low-resource languages: lnkd.in/gujhhCwm Connect with us further: * Web: votee.ai | beever.ai * Social: lnkd.in/g-9m5c84 | x.com/Votee_AI | @voteeai" target="_blank" rel="nofollow noopener">substack.com/@voteeai * Discover Votee MAGIC: Learn how to build your unique Large Language Model (LLM) tailored to your specific needs: lnkd.in/gJCJdaf4 Have a wonderful Easter! Warm regards, Votee AI team Ready to scale your AI securely? Let's talk. hello@votee.com lnkd.in/gZz-6Hab #AI #CantoneseAI #LowResourceLanguage #LargeLanguageModels #LLM

We're on South China Morning Post @SCMPNews! HOW A.I. IS HELPING TO PRESERVE CANTONESE Hong Kong-based tech firm Votee AI aims to replicate its model for other low-resource languages in Southeast Asia and Africa Preserving Cantonese has been challenging due to the dominance of Mandarin, limited learning resources and a lack of a standard written form. With a declining number of young learners, the language faces an uncertain future. Artificial intelligence (AI) – seen by some as an existential threat to humanity – may become the hope for saving the language, and many others, along with the distinct cultures they embody. This is the mission of Hong Kong-based deep-tech company Votee AI: to use large language models (LLMs) to preserve languages, especially those overlooked by tech giants. Dr. Pui-Wai (Leo) Ma, Votee’s chief scientist for the Asia-Pacific, said in an interview that AI could create living records of many languages beyond English – the primary language for most LLMs today. “While mainstream AI models excel in English, they remain ‘functionally illiterate’ for 99 per cent of the world’s languages,” Ma said, adding that this limited AI access for the billions of non-English speakers. Unlike Mandarin, English or Spanish, which have vast troves of digital text for AI training, Cantonese lacks written text that accurately reflects how it is spoken. “Cantonese is a ‘low-resource language’ because there is not much written text that we can use for training,” Ma said. "Open source is a critical path to ensuring linguistic diversity survives in the age of AI" - Dr. Leo Ma, Chief Scientist for Asia-Pacific, Votee AI Votee AI, which says it is the first company to open source a Cantonese LLM, went through the exacting challenges of recording the language spoken by 85 million people worldwide... To ensure models reflected the rich cultural nuances, a Votee AI researcher created a benchmark with peers, including those from @HKUniversity, The Education University of Hong Kong and @KyushuUniv_EN, to elevate the language proficiency and cultural knowledge, as well as reasoning and problem-solving of Cantonese LLMs in the context of Hong Kong. “While proprietary models generally outperform open-weight models, significant limitations remain in handling Cantonese-specific linguistic and cultural knowledge, highlighting the need for more targeted training data and evaluation methods,” the team wrote in a paper published by the Association for Computational Linguistics @aclmeeting last year... “Addressing the segregation of regional and linguistic knowledge is crucial for developing culturally and linguistically adaptive LLMs. This issue extends beyond Cantonese to other under-represented language communities,” they added. "We believe Al’s role is not only to create the future but also to safeguard the past" - Dr. Leo Ma, Votee AI Read the full story here: lnkd.in/gw9DB_bf Image: SCMP #AI #SovereignAI #CantoneseAI #LLM #chatbot

We're on MARKETING-INTERACTIVE! Another walled garden? As AI chatbots ramp up data collection, a familiar pattern appears to be reemerging: the creation of another walled garden—this time built not from clicks, but from users’ thought processes. In the social media era, walled gardens locked in behavioral data—likes, shares, and browsing patterns. With AI, what’s being locked in is cognitive data: users’ questions, reasoning, and vulnerabilities, said Jacky Chan, CTO of Votee AI and Beever AI. "That's a qualitatively different kind of moat. When you confide in an AI assistant about a business decision or a health concern, that insight becomes platform advantage," he added. As model capabilities commoditise, your conversations become the only real competitive differentiator these companies have. For marketers in Asia–Pacific, this carries a specific edge, he explained. “Many brands here have already navigated the fragmented super-app ecosystem—Wechat, LINE, Grab, Gojek—each with its own walled data. AI adds another layer. If your customer data lives inside one AI vendor’s ecosystem and you can’t port it out, you’ve handed over leverage you’ll never get back.” Read the full story here: lnkd.in/gp8sHw4Y Image: MARKETING-INTERACTIVE #AI #SovereignAI #CantoneseAI #LLM #chatbot

Votee AI on DotDotNews: Dr. John Whaley, Technology Advisor at Votee AI, Founder of @BuildInception, and @Stanford Academic, on the AI Agent “Pandora’s Box” In a recent feature with DotDotNews, Dr. John Whaley—Founder of Inception Studio, Technology Advisor at Votee AI, and Stanford Ph.D.—shared his insights on the rapid rise of AI agents. Referring to open‑source frameworks like OpenClaw (commonly known as “Lobster AI”), he noted: “This is like opening a Pandora’s Box that can never be closed. The industry will undoubtedly move forward very quickly.” While OpenClaw has captured global imagination by allowing anyone to build an AI agent in just a few hours, Dr. Whaley offered a candid assessment: “Frankly, OpenClaw is more like a toy.” He explained that such frameworks lack the necessary security safeguards and constraints, making them unsuitable for production environments. Yet their technical potential is undeniable—they give us a glimpse of what agentic AI can achieve. Dr. Whaley also addressed the current investment landscape, likening it to the 1990s internet revolution. “We are undoubtedly in a huge wave, and there is a bubble,” he said. In Silicon Valley, the pace of iteration is so fierce that “even models from six months ago are now obsolete.” He predicts that only companies solving high‑stakes problems with genuine differentiation and robust governance will survive the inevitable market correction. On the topic of AI replacing jobs, Dr. Whaley offered a historical perspective. He recalled the era of “human computers”—people who performed complex calculations by hand before electronic computers existed. Those jobs disappeared, but computing technology created countless new opportunities. “AI is an efficiency magnifier, not a job terminator,” he emphasized. Roles will shift from low‑level repetitive execution to higher‑level strategic judgment and oversight—areas where uniquely human capabilities like discernment and aesthetics remain irreplaceable. At Votee AI, we are committed to building the enterprise‑grade infrastructure that makes this transition both secure and sovereign. If you are seeking a business-proven solution that prioritizes: - Rigorous Governance - Intrusion Resistance - Model Agnostic Flexibility Contact our team today to secure your AI roadmap: Email: hello@votee.com Web: votee.ai/contact-page Read the full DotDotNews report: dotdotnews.com/a/202603/24/AP… Copyright: @DotdotnewsHK #VoteeAI #AIAgents #OpenClaw #EnterpriseAI #AIGovernance #SovereignAI #TechLeadership

[@SCMPNews FRONT PAGE] Dr. @joewhaley, @Votee_AI’s Technology Advisor & @Stanford Expert, on the Future of Sovereign AI The global AI narrative is shifting, and Votee AI is at the forefront of the conversation. In a definitive look at the global tech race, Votee AI's Technology Advisor—@Stanford University lecturer, 3-time cybersecurity founder, and Founder of @BuildInception, Dr. John Whaley—identifies why the "one-size-fits-all" AI model is reaching its limit. "AI is probably the most important technology that I’ve certainly seen in my lifetime and, arguably, one of the biggest shifts in the course of human history." — Dr. John Whaley The Vision: Beyond "Internet-Scale" AI As Dr. Whaley notes, foundation models are starting to converge because they rely on the same public data. The ultimate differentiator for any nation or enterprise is now proprietary data. Key insights from the feature: - The Problem with "Western" AI: Most LLMs are trained on English-centric data, leading to biases and poor performance in local contexts. - The Sovereignty Solution: Votee AI is bridging this gap for emerging economies and regulated sectors like finance, allowing them to maintain both data and cultural sovereignty. - The Strategic Mandate: Governments and organizations realize that to be at the forefront of this revolution, they must have control, visibility, and locally aware models. The Writing on the Wall For enterprises with sensitive data they cannot share, and for nations that need AI that understands their unique language and values, Sovereign AI is the only path forward. At Votee AI, we are building the infrastructure for the next generation of global productivity and security. Secure Your AI Roadmap Whether you are an enterprise looking to protect your proprietary data or a partner interested in the future of localized LLMs, let’s talk. 📩 Email us: hello@votee.com 🌐 Visit our Website: lnkd.in/gZz-6Hab 🔗 Read the full SCMP Front Page Feature: lnkd.in/g33VaG-v #VoteeAI #SCMP #SovereignAI #DrJohnWhaley #StanfordAI #TechAdvisor #DataSovereignty #IPO #GlobalTech #ArtificialIntelligence

Votee AI Featured on DotDotNews: Bridging the Security Gap in Agentic AI Dr. Pui-Wai (Leo) Ma, Chief Scientist (APAC) at Votee AI, recently highlighted that the core risks of current frameworks like "OpenClaw" stem from their inherent design. Most intelligent agents today lack robust intrusion resistance, allowing attackers to embed malicious instructions within routine data—such as emails and webpages—to trigger unauthorized actions. Dr. Ma further notes that over-reliance on a single model leads to "correlation failures," while unvetted supply chains and a lack of pre-deployment audits leave systems highly vulnerable. From Risk to Resilience: The Votee Solution At Votee AI, we believe "good enough" AI is a strategic liability. We provide Enterprise Grade Sovereign AI designed for high-compliance environments where security and governance are non-negotiable. If you are seeking a business-proven solution that prioritizes: - Rigorous Governance - Intrusion Resistance - Model Agnostic Flexibility Contact our team today to secure your AI roadmap: 📩 hello@votee.com 🌐 lnkd.in/gZz-6Hab 🔗 Read the full DotDotNews report: lnkd.in/guCrc8xj #VoteeAI #AISecurity #VPTIndex #AIValuation #AutonomousAgents #EnterpriseAI #TechLeadership #IPOBound #AIGovernance #CantoneseAI #SovereignAI

Building a specialized Cantonese LLM and scaling agentic AI for enterprise clients requires a robust, flexible foundation. We are thrilled to partner with MongoDB to make this happen. As Gabriel Woo, Regional Vice President, Greater China Region at @MongoDB, notes, MongoDB acts as the "shared brain" that allows our AI platform to deliver real-time, global market intelligence across the banking, telco, and retail sectors. Thank you to Gabriel and the entire MongoDB team for the incredible support and feature, and a massive shoutout to our very own Dr.Pui-Wai (Leo) Ma, Chief Scientist, APAC, for leading this technical vision! Check out our story in the video below. youtube.com/shorts/IeJMMXD…