Mr. Md. Shamsul Arefin, Secretary, Information and Communication Technology (ICT) Division; Dr. Syeda Naushin Parnini, Additional Secretary, Ministry of Science and Technology; Mr. Mirza Salman Hossain Beg, Co-Founder & COO, 10 Minute School; Mr. A. S. M. Shamim Reza, Founder & Chief of Research, The Team Phoenix; Dr. Pasha, Senior Researcher / Nuclear Medicine & Healthcare Specialist; Dr. Shamsul Alam, Database In-Charge, CMH Nursing College; Representative, Information and Communication Department, Bangladesh Bank; Representative, Ministry of Foreign Affairs (Director General); Representatives from UCL Digital Lab, NHS AI Implementation Team, and TMSS; S.M. Sarfaraj, CEO, TRIAD THE CYBER; Luca Licata, CTO, TULIPTECH; Haider Anwar, CEO, Digital Bridge Lab; Nabil Chowdhury, COO, Digital Bridge Lab; Dr. Md. Matiur Rahman, DED, TMSS Bangladesh; Sayed Ahamad, Trade Analyst, Einbany; Sharif Ahmed, Senior Reporter, Press Information Department (PID); Imtiaz Islam, Business Development, Consultive; Muhammad Monirul Islam, Deputy Director, TMSS, Bogura BD; ASM Shamim Reza, Managing Director, The Team Phoenix; Mirza Salman Hossain Beg, Founder, CEO, iKori; Dr. Md. Nazrul Islam Khan, Chief Scientific Officer, BAEC; Dr. Mohammad Anwar ul-Ajim, Principle Scientific Officer, BAEC; Alam Been Azam, LEAD ORGANISER, AI TINKERS DHAKA; Dr. M.A. Talukder, Professor, BUET; Dr. M.S. Bashar, PSO, BCSIR; Kamer Bin Siddique, Faculty, TMSS Nursing College; Mustak, Co-Founden MD, Puku AI; Galib, Student, BAUST; Esmail, Student, SUBAT; Debashis Dhali, Student, Jahangirnagar University; Ismail Zabihullah, PMO; MD. Mamunur Rashid Bhiyan, Secretary; Dr. Salma, JS, MOST; Debbarto Kumar JSTU, Students, Students, JSTV; Mohammad Hassan, Assistant Professor, CSE, JSTU; Shabbar Mustafa, Advisor, Penta Global; Dr. Syed, MOFA; Dr. Syeda Parvin, Additional Director, MOST; Fasbeer Eskander, Founder Director, The Front Page.

The Day 02 Roundtable at the Bangladesh Innovation Fair-2026 addressed the structural and regulatory prerequisites for establishing a National AI Platform. Emphasizing the transition from the largely unrealized 2019–2020 AI Strategy to the Draft National AI Policy (2026–2030), the discussion centered on whether Bangladesh should pursue capital-intensive foundation models or focus on shared high-performance computing, practical sector-specific deployments, and open data access. Key imperatives identified included democratizing access to enterprise-grade GPU compute to curb foreign cloud dependency, enforcing tiered data governance to protect citizen PII and sensitive healthcare records, structuring R&D tax incentives for enterprise collaboration, and addressing passive infrastructure bottlenecks such as power and cooling water consumption.

• Mr. Md. Ismail Zabihullah (Chief Guest): Argued that AI must be treated primarily as a national productivity and competitiveness agenda rather than just another software technology. Emphasized that raw human talent alone does not create national capability unless systematically converted into research, enterprise, and export-oriented employment. Warned against protecting operational inefficiencies for the sake of preserving low-productivity jobs, while stressing the need to upskill the workforce for technology-augmented roles.
• Mr. Md. Shamsul Arefin (Secretary, ICT Division): Affirmed that the current democratic government is actively revising the draft AI policy under a specialized committee led by the Hon'ble ICT Adviser. Noted that developing an indigenous Bangla LLM and fine-tuning existing models are not mutually exclusive; the state must balance immediate citizen-centric problem-solving with foundational capacity-building.
• Dr. Syeda Naushin Parnini (Additional Secretary, MoST): Positioned AI as fundamental infrastructure akin to electricity or the internet. Emphasized the urgency of preparing the nation's youth (Gen-Z) for the shift toward the Fifth Industrial Revolution and developing localized LLMs catering to Bangladesh's rich linguistic diversity (e.g., regional dialects like Chittagonian, Noakhali, and Sylheti) for critical services like rural telemedicine.
• Mr. Mirza Salman Hossain Beg (COO, 10 Minute School): Provided a reality check on building a foundation "Bangla LLM" from scratch, citing immense capital, engineering, and compute requirements. Recommended following the model of nations like Qatar and Oman, which subsidize compute and model access for citizens to build applied products rather than draining national reserves on foundational training.
• Dr. Moontaha (Expert, AI & Quantum Mechanics): Noted an inverted SWOT analysis in Bangladesh: data depth and interoperability, which should be national strengths, have become systemic weaknesses due to institutional silos. Argued that AI is cognitive replication rather than an off-the-shelf consumable, warning that the country has an implementation window of at most 18 months before global advancements outpace domestic capabilities.
• Mr. A. S. M. Shamim Reza (The Team Phoenix): Emphasized that human psychology and digital literacy are the foundational prerequisites of AI adoption. Pointed out that the absence of standard disclaimers for AI-generated content and rising AI-driven fraud (e.g., automated WhatsApp traffic fine scams) demonstrate acute gaps in responsible usage and cybersecurity readiness.

• Foundational "Bangla LLM" vs. Applied Fine-Tuning: The strategic debate on whether to invest massive capital into pre-training an indigenous foundational LLM or leverage shared GPU infrastructure and open-weight models to address immediate vertical challenges in healthcare, logistics, and agriculture.
• Compute Shortages and Capital Outflow: The severe lack of accessible local enterprise GPU clusters forces local startups and enterprises to pay substantial foreign currency reserves to foreign cloud providers (e.g., in Thailand and Singapore).
• Data Sovereignty, Silos, and Tiered Privacy: Centralizing fragmented public databases (e.g., millions of consular records at MOFA, clinical datasets at public hospitals) under secure, standardized data repositories with strict tiered protections for citizen Personally Identifiable Information (PII).
• Environmental & Resource Impact of AI: Assessing the high electric power demands and massive cooling water footprints required to run hyperscale enterprise data centers in a water-stressed, developing economy.
• Academic Reform vs. AI Prohibitions: Rejecting outright bans on generative AI in schools and universities in favor of human-AI collaboration, faculty upskilling, and dynamic assessment models.
• Triple Helix Collaboration & Sandboxes: Structuring institutional mechanisms (e.g., regulatory sandboxes, R&D tax credits) allowing academia, private industry, and state agencies to co-develop and safely test prototypes.

• Compute & GPU Deficits: Local researchers and startups cannot access high-performance GPUs, crippling their ability to run local inference or train domestic models without resorting to expensive overseas cloud infrastructure.
• Data Fragmentation and Absence of Registries: The lack of standardized national health data repositories (such as cancer or trauma registries) forces international bodies like the WHO to rely on synthetic proxy data for Bangladesh.
• Capital Flight via Cloud Rents: Domestic AI companies are forced to bleed thousands of dollars monthly in foreign currency to host compute workloads abroad due to the absence of domestic infrastructure.
• Institutional Siloing and Bureaucracy: Large volumes of valuable administrative and public data remain trapped within uncoordinated ministry silos; prolonged bureaucratic processes delay data access for domestic developers.
• Algorithmic Injustice & False Positives: Automated enforcement systems (such as traffic cameras ticketing owners solely via license plates without verifying driver identity) expose citizens to administrative harassment with zero state liability or compensation mechanisms.
• Environmental Sustainability Oversight: High-performance data center proposals frequently overlook the immense water consumption (up to 1 liter per heavy prompt cooling) and power grid loads necessary to support dense clusters.

• Shared Public-Private GPU Compute Clusters: Incorporate enterprise-grade GPU clusters into state projects to provide subsidized compute time for researchers, students, and startups, bypassing expensive private capex duplication.
• Open-Source Public Data Approach: Emulate Malaysia's Department of Statistics by establishing an expedited 5-day audit committee to anonymize and publicly release non-sensitive demographic and economic datasets via open-source APIs.
• Tiered Data Governance Architecture: Adopt a UK NHS-style tiered framework where general administrative data is open, while clinical patient data and PII are protected under security protocols second only to national defense.
• Regulatory Research Sandboxes: Establish safe-harbor sandboxes (similar to fintech sandboxes used by central banks) to allow researchers and startups to test algorithms, fail safely, and refine models prior to commercial scaling.
• Fiscal Incentives for Industrial AI R&D: Introduce R&D tax credits and matching grants to incentivize telecommunications providers, banks, and large corporations to share infrastructure and invest in university labs.
• Algorithmic Redundancy (Triangulation): Mandate a minimum of three parallel data inputs (e.g., dual cameras and cross-verification feeds) before an automated civic penalty or legal action can be triggered against a citizen.

• United Kingdom (NHS): Utilized pre-existing GDPR and cybersecurity baselines to establish a tiered data sensitivity model, successfully accelerating AI deployment while strictly safeguarding sensitive clinical records.
• Malaysia (DOSM): The Department of Statistics opened up demographic and economic datasets via open-source dashboards, enabling independent developers and data scientists to build nationwide pandemic response tools with zero state procurement overhead.
• United States (Frontier Landscape & K-12): Silicon Valley's rapid model iteration vs. reactionary public school bans (e.g., NYC public school bans and student suspensions), illustrating the perils of prohibition over integration.
• Middle East (Qatar & Oman): Governments subsidizing citizens' access to premier foundation models to stimulate grassroots product development.
• Israel: A small economy that successfully built and scaled over 2,500 applied AI companies, generating billions in technology export revenues.

• Dhaka Metropolitan Police (DMP) AI Cameras: Live traffic violation monitoring on Dhaka's streets and 490 highway pillars along the Dhaka–Chattogram corridor, highlighting both automation leaps and license-plate ticketing conflicts.
• Ministry of Foreign Affairs (MOFA): Successfully engineered a comprehensive consular ecosystem processing 12 million document attestations annually on a lean 30,000 USD budget.
• Capital Outflows (Thailand Cloud Hosting): A local tech startup paying 32,000 USD monthly to Huawei Cloud in Thailand due to the absence of domestic high-performance GPU hosting.
• Underutilized Public Facilities: The Bhashani Novo Theatre and the Tier-IV National Data Center identified as public assets with substantial underutilized capacity that could be repurposed for shared compute and innovation hubs.

• ICT Division & NBR: Establish an enforceable R&D tax credit policy to encourage telecommunications operators, commercial banks, and enterprise tech firms to donate GPU compute time and open anonymized datasets to researchers.
• Cabinet Division & Law Ministry: Enact dedicated provisions under the proposed Data Governance Act establishing state liability, fair compensation, and dispute resolution for citizens impacted by algorithmic or automated enforcement errors.
• Ministry of Science and Technology: Expedite the physical and institutional establishment of the National AI Institute / Center of Excellence to act as an agile, permanent body capable of updating policies every 6–12 months.
• Ministry of Health and Family Welfare: Construct an interoperable, privacy-compliant National Health Data Exchange, establishing sovereign registries for non-communicable diseases (oncology, cardiology) to support applied medical AI research.
• Ministry of Education & UGC: Eliminate institutional bans on generative AI; mandate human-AI collaborative workflows, ethical prompt engineering, and digital literacy across secondary and tertiary academic curricula.

• 180 million: Total population of Bangladesh, characterized by a disproportionately high density of engineers relative to national economic footprint.
• 12 million: Number of document certifications and attestations processed annually in a single silo by the Ministry of Foreign Affairs.
• USD 32,000/month: Foreign currency expenditure incurred by a single local tech company to rent GPU computing infrastructure from Huawei in Thailand.
• 1 liter per prompt: Estimated cooling water consumed across data center infrastructure per heavy generative query prompt (highlighted by Smart Bill).
• 750 to 50: Over 750 innovation proposals submitted to the Ministry of Science and Technology, from which 50 high-potential prototypes were selected for exhibition and acceleration.
• 1,500 BDT: The black-market price for unauthorized Call Detail Records (CDRs) and phone records in Bangladesh, illustrating severe data sanitization and baseline privacy failures.
• 6 to 7 months: Official target timeline committed by the ICT Division to deploy centralized, shareable high-performance computing capacity under approved state projects.

• Innovation Hub Launch: Full operationalization of the innovationhubbd.com platform launched by the Ministry of Science and Technology to curate, protect, and fund the 50 selected fair prototypes and link them with investors.
• Centralized Compute Delivery: The ICT Division's formal commitment to make shared enterprise computing and GPU clusters operational for universities and private enterprises within 6 to 7 months under two approved projects.
• Legislative Suite Enactment: Fast-tracking the revision of the Cyber Security Act, the tabling of the Personal Data Protection Act, and the passage of the Data Governance Act.
• Establishment of Regulatory Authority: Formalizing the creation of the National Data Governance Authority under the Prime Minister's Office to regulate cross-agency data exchange, interoperability standards, and privacy compliance.
• National AI Summit: The Ministry of Science and Technology announced plans to convene a high-level National AI Summit to review time-bound action plans and formalize the permanent National AI Institute.

• Moving Beyond "Cheap Labor": National economic strategy must break away from reliance on low-cost, low-productivity labor in the RMG sector; AI must be leveraged to drive high-margin value creation ahead of LDC graduation.
• Inter-Ministerial Harmonization: Multiple ministries (ICT Division, Ministry of Science and Technology, Ministry of Foreign Affairs, Ministry of Public Administration) must coordinate their technology initiatives to eliminate redundant capex investments and prevent overlapping jurisdictions.
• Sovereignty vs. Vendor Lock-In: Reliance on proprietary foreign APIs (OpenAI, Anthropic, Google) without building domestic data pipelines leaves national critical infrastructure vulnerable to unilateral external policy shifts and subscription price hikes.

1. Focusing on Practical Productivity Over Hype: Bangladesh must avoid the temptation of chasing vanity projects or premature foundational model development. AI must be implemented where it delivers tangible macroeconomic gains: factory error reduction, precision agricultural yields, optimized urban transit, and rapid citizen service delivery.
2. Developing the "National Platform for AI Innovation": Rather than establishing a bloated bureaucratic body, the state will create a nimble operational clearinghouse connecting government problem statements, academic research prototypes, private enterprise commercialization, and venture scale.
3. Pivoting from Exploitation to Capability: Transforming Bangladesh's competitive positioning from "cheap labor" to a high-productivity, technologically capable economy by making technology and human workforce skills reciprocal forces.
4. Agile Execution ("Pilot, Evaluate, and Scale"): Refusing to wait indefinitely for the "perfect" policy or total infrastructural readiness; the state and private sector must collaborate immediately across live pilot sandboxes, learning iteratively to power the nation's future.