On October 7, 2026, aivancity hosted Professor Subhankar Dhar from San José State University for a three-hour workshop titled "AI Workshop: From Smart Recommendations to Real-World Innovation."
Designed for Master's students, the morning session progressed from a broad overview of the ecosystem to a live product demonstration, followed by a hands-on innovation challenge.
Professor Dhar brings a rare blend of perspectives. He is a professor at SJSU, a senior member of the IEEE, and editor-in-chief of the IJBDCN. His research focuses on AI in education, recommendation systems, trustworthy AI, big data, and smart cities, with work funded by the NSF and projects conducted with the City of San José. He also mentors startups and serves on the judging panel for the Silicon Valley Business Plan Competition, making him an ideal guide for a session bridging the gap between academia and entrepreneurship.
The session covered four main topics:
- The Silicon Valley innovation ecosystem and how AI is changing the game for startups
- AI in education, featuring a live demonstration of the agent-based platform myAcademic
- Recommendation systems in practice, followed by a practice quiz
- An innovation challenge on AI in education, concluding with five-minute team pitches
Silicon Valley: A Network, Not Just a Place
Professor Dhar opened the session with a key idea: AI accelerates innovation; ecosystems make it decisive. The numbers show why this moment is important. According to the Stanford HAI 2026 AI Index:
Silicon Valley's strength rests on six mutually reinforcing ingredients: talent, knowledge, capital, infrastructure, markets, and a culture that tolerates failure. Together, they drive a virtuous cycle. Research fuels startups, startups attract capital, capital enables scaling, and returns are reinvested in new founders. AI accelerates each cycle by reducing the cost of prototyping.
How AI Is Changing the Startup Equation
Natural language programming, agents, and reusable APIs enable smaller teams to build a credible prototype and test more hypotheses each month. The flip side: new risks, such as model dependency, variable inference costs, and weak differentiation. Sustainable value continues to come from domain expertise, proprietary workflows, trust, distribution, or a genuine data advantage.
The students also explored the AI technology stack where startups compete, from computing and power at the foundation, through data, models, and orchestration, all the way to vertical applications in education and healthcare. Professor Dhar showed how OpenAI, Google, Anthropic, and NVIDIA each run startup programs that combine access to technology, credits, mentoring, and market connections.
The Founder's Guide
An AI startup always succeeds by better solving a painful problem. The guide shared with students:
- Start with a workflow: observe users, constraints, and the cost of failure.
- Validate the entry point: deliver a result that is measurably faster, less expensive, or better.
- Design for reliability: through evaluation, factual grounding, permissions, and escalation to a human.
- Build a defensive advantage: in the workflow, data, trust, or distribution.
- Scale responsibly: by tracking unit economics, model risk, and customer outcomes.
As one slide summarized, a demo shows what is possible, but a company delivers a reproducible and trustworthy result. Responsible AI, he argued, is not a compliance layer but a component of product-market fit.
Responsible AI is not a compliance layer, but a component of product-market fit.
Prof. Subhankar Dhar
Universities: Active Nodes in the Innovation Network
Where do schools like aivancity fit in? Professor Dhar's answer: universities should not observe the ecosystem from the sidelines but should be active nodes within it. He described five roles:
- Teach: combine mastery of AI, depth in a specific field, and entrepreneurial judgment.
- Build: organize studios, capstone projects, hackathons, and innovation challenges.
- Translate: transform research into products through partnerships with industry and startups.
- Govern: develop capabilities in evaluation, ethics, and public policy.
- Include: expand access to tools, mentors, computing power, and networks.
He also discussed concrete forms of academic collaboration between institutions: guest lectures that evolve into joint modules and courses, shared research themes such as trustworthy AI, agent-based AI, and recommendation systems, as well as faculty and student exchanges through visiting professor positions, internships, and mentoring. The next wave of AI, he noted, is shifting from generic chatbots to domain-specific applications in sectors such as education, healthcare, and law, where universities' expertise and credibility are real assets.
AI in Education: A Live Look at myAcademic
The second part of the morning shifted from ecosystems to a concrete product. Professor Dhar presented myAcademic, an intelligent learning platform based on agent-based AI, developed by Prof. Dhar, that guides students from a generic learning management system (LMS) to professional success, and demonstrated it live, including AI feedback on a resume for career opportunities.
The starting point is simple. Traditional LMS platforms organize learning: grade books, content repositories, manual analyses. The next step is to understand, predict, recommend, and take action. myAcademic adds a layer of agent-based AI on top of an institution's existing LMS rather than replacing it.
Four Key Components
Two Perspectives, One Loop
For students, the platform transforms grade data into a continuous success loop: sync the latest course data, analyze gaps, predict outcomes, recommend actions, then study and make progress. It answers questions such as "What should I focus on this week for my next exam?" or "Can I turn in my assignment late?" In addition, it allows students to upload their resumes and recommends relevant job openings nearby that match the student's expertise.
For instructors, AI is involved throughout the entire teaching cycle, from curriculum design to grading, including identifying students who need support. Crucially, the instructor retains control through human-in-the-loop validation, transparent recommendations, and permissions defined by the institution. When a student is flagged, the platform explains why.
From Classes to Careers
The most forward-looking part of the demonstration extended the journey beyond the classroom. A career agent builds a skills profile based on courses, links course learning to specific skills, identifies gaps, and recommends courses, projects, and internships. A student might ask, "I want to become a data analyst. Based on my courses and performance, what should I focus on next?"
Safeguards are essential here. Career suggestions remain advisory, explainable, and under the student's control; they are based on approved academic data; and they never infer protected characteristics or make high-stakes decisions on their own.
Student Innovation Challenge
After a session on recommendation systems in practice and a short practice quiz, students put theory into action during the AI in Education Innovation Challenge. The topic:
The problem statement addressed a challenge familiar to every programming instructor. Instructors need a better way to grade programming assignments, as manually checking every solution and every edge case is impossible. Teams were asked to design an AI-based platform that automatically detects errors and suggests corrections.
Five roles per team
The students formed teams of five, with each member bringing a different perspective to ensure that nothing essential was overlooked:
- Problem Lead: defines the target user, their needs, and the limitations of the current approach.
- AI Lead: proposes the AI task, model, or agent-based approach.
- Data Lead: identifies data requirements, quality limitations, and access constraints.
- Responsible AI Lead: assesses privacy, fairness, security, exclusion, and misuse.
- Product and Pitch Lead: shapes the workflow, value proposition, and final presentation.
A Ten-Step Innovation Framework
The teams worked from a framework covering the user, the problem, the current approach, AI capabilities, and data, followed by the workflow, the human role, value, success metrics, and risk. The question of the human role stood out: where should judgment, consent, or approval remain the responsibility of a person?
Each team then gave a five-minute pitch using three slides: the problem and why current approaches are insufficient; the AI solution, its data, and its design with humans in the loop; and finally, the benefits, success metrics, risks, and safeguards.
AI accelerates innovation; ecosystems make it decisive.
Prof. Subhankar Dhar
