AI Ketchup 🍅 | Your Business's Secret Sauce – Détails, épisodes et analyse

In this episode, we dive deep into the world of knowledge graphs and organizational change with Vadym Safronov, a Lead Data Scientist at Nielsen IQ and veteran of enterprise transformations. From building actual ketchup factories to architecting complex data systems, Vadym shares fascinating insights on how knowledge graphs can revolutionize enterprise operations and drive successful organizational change.

TOPICS DISCUSSED:

1. Knowledge Graphs Fundamentals
Vadym breaks down the concept of knowledge graphs through practical examples, explaining how simple subject-predicate-object relationships can be used to build complex knowledge systems. He illustrates how these structures can be enhanced with neural networks to predict patterns and relationships.

2. Enterprise Transformation
The discussion explores how knowledge graphs can map organizational structures, processes, and relationships to drive successful change initiatives. Vadym shares insights from his experience at Nestle and other enterprises on identifying effective change agents through network analysis.

3. The Science of Change Management
We explore the fascinating research behind successful organizational change, including the importance of network topology in selecting change agents and why traditional approaches often fail. Vadym explains why focusing on early adopters rather than innovators leads to more successful transformations.

4. Combining Knowledge Graphs with Generative AI
The conversation examines how enterprises can leverage both knowledge graphs and large language models to create more reliable and factual AI systems, using the metaphor of having both a master librarian and universal interpreter at your disposal.

INSIGHTS:

- The power of structural patterns in predicting organizational behavior
- Why three out of four IT interventions fail due to non-technical reasons
- The importance of network topology in selecting change agents
- How knowledge graphs can help combat misinformation
- Why focusing on early adopters rather than innovators leads to more successful change initiatives

TOOLS AND TECHNOLOGIES MENTIONED:

- SAP CRM
- DBpedia
-The Network Secrets of Great Change Agents
-Known hoaxes on Wikipedia
-Wikispeedia game

CONTACT INFO:

-Vadym Safronov

CHAPTERS
00:00 Introduction and Background
03:15 From Ketchup Factories to Data Science
07:30 Evolution of Graph Applications
12:45 Understanding Knowledge Graphs
18:20 Combining Knowledge Graphs with Generative AI
23:40 Enterprise Change Management
31:15 Network Analysis for Change Agents
38:50 Rogers' Innovation Adoption Theory
45:30 Knowledge Graphs and Misinformation

Follow AI Ketchup for bi-weekly stories of AI builders and founders turning ideas into successful tech products.

In this episode, we explore the evolution of AI in product management and the crucial balance between automation and human judgment with Jorge Alcantara, a founder of Zentrik.ai and a veteran in AI implementation and product development. Jorge shares his journey from early chatbot development to founding Zentrik, offering unique insights into the future of product management in the AI era.

TOPICS DISCUSSED:

1. The Evolution of AI Implementation
Jorge shares his experience with early chatbot deployments and the transition from rule-based to generative AI systems. He emphasizes how the focus has shifted from pure automation to augmenting human capabilities and understanding user needs through Human-in-the-Loop training mechanisms.

2. The PM Paradox
We explore the current challenges in product management, where PMs often become "Jira janitors" instead of focusing on high-value activities like user research and strategic planning. Jorge explains how AI can help rebalance PM workflows and why companies need to rethink their approach to product management.

3. Human-Centric AI Development
The conversation delves into the importance of maintaining human judgment in AI solutions, particularly in product management. Jorge emphasizes that while AI can automate routine tasks, the real value comes from freeing PMs to focus on empathy, user understanding, and strategic thinking.

4. The Future of Product Management
Jorge presents his vision for how AI tools should evolve to support product managers, highlighting the importance of specialized solutions over generic AI tools. He discusses how proper AI implementation can help companies build better products by enabling PMs to spend more time on high-value activities.

INSIGHTS:

1. Product management is becoming the skill of the future as development gets commoditized.
2. The importance of freeing PMs from routine tasks to focus on user research and strategic thinking.
3. Why generic AI tools only solve 10% of PM-specific challenges.
4. The need for specialized AI solutions in product management.
5. The value of human judgment and empathy in product development.

TOOLS AND TECHNOLOGIES MENTIONED:

- Zentrik.ai
- ChatGPT
- Canvas
- ChatPRD
- Wiser

CONTACT INFO:

- LinkedIn: Jorge Alcantara
- Email: jorge@zentrik.ai

CHAPTERS
00:00 The Evolution of Chatbots and AI Technology
01:18 How Chatbots Started and Fears around AI
05:45 Bringing Human Emotionality to Machines
08:00 Rewarding Human-in-the-Loop
14:25 Disturbing Trends & Product Management Paradox
17:35 AI's Impact on Product Management Tasks
20:39 How ML can help with PMs' multidisciplinarity
25:45 The Main Pain Mentioned within 200+ interviews
26:26 Challenges in Product Management Documentation
28:02 AI Tools for Product Management Efficiency
32:09 ChatGPT's Canvas and Realtime API for PMs
34:28 Communicating AI Benefits to Executives
39:12 PM's Mastery in Times of Commoditized Code
40:57 Empathy in Product Management
45:40 Personal Reflections on Work and Life Balance

Follow us at AI Ketchup for bi-weekly stories of AI builders and founders turning ideas into successful tech products.

In this episode, we dive into the fascinating world of quantum computing with Hila, a quantum computing researcher at Siemens focusing on hardware-software co-design. From her unexpected journey from aspiring surgeon to quantum computing expert, Hila brings unique insights into the future of this transformative technology and its real-world applications.

TOPICS DISCUSSED:

1. Quantum Computing Fundamentals
Clear explanation of quantum computing principles using analogies (coins, ripples in water). Comparison between classical and quantum computers using the candle vs. lightbulb metaphor. Detailed breakdown of key quantum properties: superposition, interference, and entanglement. Discussion of how quantum computers complement rather than replace classical systems.

2. Current Challenges and Solutions
Deep dive into error correction challenges in quantum systems. Explanation of physical vs. logical qubits. Analysis of different quantum hardware approaches (superconducting vs. ion trap systems). Discussion of the NISQ (Noisy Intermediate Scale Quantum) era and its implications.

3. Technical Implementation
Hardware-software co-design considerations. Discussion of different quantum hardware technologies. Integration with classical computing systems. Future outlook for quantum computing development.

4. Practical Applications
Material science and molecular simulation. Drug discovery and personalized medicine. Supply chain optimization and logistics. Climate modeling and environmental applications. Quantum machine learning potential.

5. Social Impact and Responsibility
Emphasis on ethical guidelines and regulations. Importance of transparency in quantum research. Need for collaborative approach across disciplines. Focus on making quantum computing accessible and understandable.

INSIGHTS:

1. Quantum computers are best suited for specific tasks rather than general-purpose computing.
2. Error correction remains a major challenge requiring multiple physical qubits per logical qubit.
3. Different quantum hardware architectures offer various trade-offs for different applications.
4. The field requires early consideration of ethical implications and responsible development.

TOOLS AND TECHNOLOGIES MENTIONED:

- UN announcement of 2025 as a year of quantum science and technology
- Google Willow: Google Willow Quantum Chip
- IBM Quantum Systems
- Google's Error-Corrected Quantum Computer Prototype
- Quantum Hardware Platforms (Superconducting, Ion Trap)
- Einstein–Podolsky–Rosen (EPR) paradox

CONTACT INFO:

- LinkedIn: Hila Safi

CHAPTERS
00:00 From Hospitals into Quantum Computing
04:29 ELI5: Quantum Computing Walkthrough
07:58 Classical VS Quantum Computers
11:20 Is the Future of Machine Learning in Quantum?
15:30 Why is Error Correction Necessary?
19:08 Good Enough Number of Qubits
23:58 Unexpected about Cryptography and Solving Travelling Salesman
30:17 Unclarity, Perseverance and Society
33:48 The Societal Impact and Ethical Considerations of Quantum Technology
37:39 Reproducibility in Science

Follow us at AI Ketchup for bi-weekly stories of AI builders and founders turning ideas into successful tech products.

In this episode, we dive deep into the world of local AI models and their creative applications with Shahbaz Mansahia, an ML engineer, an IEEE author, researcher and advocate for democratizing AI technology. Shahbaz shares his unique perspective on running AI models locally, making AI accessible to art students, and using technology to address representation gaps in art.

TOPICS DISCUSSED:

1. Local Models
Shahbaz explains how running AI models locally offers freedom from service providers while maintaining similar capabilities through quantization. He discusses the trade-offs between model size and performance, sharing insights about the future of 4-bit quantization and its potential for mobile AI deployment.

2. AI in Education
Making AI technology accessible to students presents unique challenges. Shahbaz discusses how open-source alternatives to commercial AI services can democratize access while maintaining quality, emphasizing the practical applications in academic environments.

3. AI and Artistic Creation
Rather than viewing AI as a threat to creativity, Shahbaz presents it as a tool for enhancing artistic workflows and democratizing expression. He shares his experience working with art students and using AI to address historical representation gaps in art.

4. Technical Implementation
The conversation covers practical aspects of local model deployment, including multimodality challenges and the evolution of hardware requirements. Shahbaz provides insights into the future of CPU vs. GPU computing for AI and the development of inference-optimized hardware.

INSIGHTS:

1. One can break dependency from any model provider with local models.
2. Foundation models work like the internet in your pocket.
3. Local deployment enables better privacy control for sensitive data. 4. We can amplify bias to extend the perceptions of artworks.

TOOLS AND TECHNOLOGIES MENTIONED:

- LM Studio
- Hunyuan Text-to-Video Model
- Comfy UI
- Dreambooth

CONTACT INFO:

- LinkedIn: Shahbaz Mansahia
- Email: shahbazsinghmansahia@gmail.com

CHAPTERS
00:00 Intro into Shahbaz's Background
02:40 Local Models and Their Advantages
05:57 Quantization and Model Performance
10:01 Practical Applications of Local Models
14:53 Multimodality and Future Developments
18:07 The Role of CPUs and GPUs in AI
20:57 AI in Art: Creativity vs. Automation
25:56 Bias in AI and Art Representation
32:02 Ethics of AI in Art and Representation

Follow us at AI Ketchup for bi-weekly stories of AI builders and founders turning ideas into successful tech products.

AI Ketchup - Catching up on the WHY behind tech products!

In this podcast, we're re-engineering human-centered product creation, focusing on AI solutions that matter. Think of this as your opportunity to step into the shoes of people solving real-world problems.

In the early days of cars, building engines wasn't easy, but it was a problem with a clear direction—unlike the challenge of defining traffic rules and places for roadways. Today, AI's technical barriers are falling, but the real focus is on what to build and why. Seven years ago, when I started exploring teh area of data science and machine learning, AI was confined to labs and very progressive universities. Now, with democratized tools, the HOW is simpler—it's time to master the WHY.

In each episode, founders share their decision-making process and builders reveal their methods, while you join the conversation thinking through how to solve for specific human needs. Join us twice a month—let's build what truly matters.

In this episode, we dive into AI development strategies with Nico Finelli, Founding GTM at Vellum.ai. Drawing from his experience with autonomous vehicles at M City and his work at Weights & Biases, Nico shares valuable insights on building AI agents through an incremental, test-driven approach.

TOPICS DISCUSSED:

1. Incremental AI Development
Nico draws parallels between autonomous vehicle development and AI agent building, emphasizing the importance of starting with constrained tasks and gradually expanding capabilities. He illustrates how this approach can be applied to various domains, from legal to healthcare applications.

2. AI Implementation in Healthcare
Through case studies like DeepScribe, Nico demonstrates how AI can enhance healthcare workflows while maintaining human oversight. He discusses practical approaches to evaluation and implementation in sensitive domains.

3. State of AI Development
Drawing from Vellum's comprehensive developer survey, Nico shares insights about AI implementation challenges, highlighting that only 25% of teams successfully reach production, and discusses strategies for improving these outcomes.

4. Evaluation Frameworks
The conversation explores practical approaches to evaluating AI systems, emphasizing the importance of structured testing and feedback loops in development cycles.

INSIGHTS:

- The value of constraining AI problems to build competency gradually

- Why human oversight remains crucial in sensitive AI applications

- The importance of robust evaluation frameworks in AI development

- How implicit and explicit feedback shapes AI system improvement

- The role of domain experts in defining AI system constraints

TOOLS AND TECHNOLOGIES MENTIONED:

- Vellum SDK

- Electronic Health Records (EHR) Systems

- LLM Orchestration Platforms

- SOAP Notes

- Test-Driven Development Frameworks

USEFUL LINKS:

- State of AI Report 2025:

- Vellum Case Studies:

- Vellum SDK Documentation:

CONTACT INFO:

- LinkedIn:

- Email: nico@vellum.ai

CHAPTERS

00:00 Path to a Founding GTM at Vellum.ai

07:03 Insights from Weights and Biases and Vellum.ai

14:05 Building AI Agents: Lessons from Autonomous Vehicles

21:44 The State of AI: Insights from Developer Reports

28:02 Vellum SDK: Enhancing AI Development and Evaluation

34:34 Encouragement for Aspiring AI Builders

In this episode, we explore the intersection of AI security, vector databases, and career transformation with Thierry Damiba, Developer Advocate at Qdrant. From his experience securing sensitive government applications to pioneering vector database implementations, Thierry shares valuable insights on building secure AI systems and navigating technological change.

TOPICS DISCUSSED:

1. Vector Databases and AI Security
Thierry explains how vector databases have become the ideal data management tool for AI applications, discussing their role in securing sensitive data and implementing effective access controls. He shares practical approaches to preventing hallucinations and data leakage in AI systems.

2. Security in the Age of AI Agents
The conversation delves into the implications of AI agents for security, exploring both the challenges and opportunities they present. Thierry discusses how automation is actually increasing the value of deep technical understanding while making technology more accessible.

3. HNSW Algorithm and Vector Search
Through an engaging library analogy, Thierry breaks down the complexities of the HNSW algorithm, explaining how it enables efficient vector search at scale and why this matters for modern AI applications.

4. Career Evolution in the AI Era
The discussion examines the changing landscape of technical careers, with insights on adapting to automation and finding fulfillment in technological work. Thierry shares personal experiences of prioritizing passion over immediate financial gain.

INSIGHTS:

- The dual role of AI agents in both creating and preventing security vulnerabilities
- Why open source contributes to better security in AI systems
- The importance of implementing both API-level and data-level security measures
- How automation is transforming the value proposition of technical skills
- The significance of pursuing passion in career choices during technological transformation

TOOLS AND TECHNOLOGIES MENTIONED:

- HNSW Algorithm
- JWT (JavaScript Web Tokens)
- GPU Indexing for Vector Databases
- Small Language Models (SLMs)
- Qdrant

CONTACT INFO:

- Twitter: @ptdamiba
- Email: td@qdrant.com
- Discord: Qdrant Community Channel

CHAPTERS
00:00 The Rise of Vector Databases
01:54 Security in AI Applications
05:14 Guardrails for AI Systems
08:13 Jailbreaking and Input Validation
09:54 AI Agents: Opportunities and Risks
16:53 The Future of Work and Automation
25:45 GPU Indexing and Application Development

Can the marriage of philosophy and technology create a more ethical AI future? Discover how Jean Arnaud, a philosopher turned AI innovator, is pioneering a revolutionary approach to responsible AI development through his concept of "digital renaissance." Jean shares his fascinating journey from teaching philosophy in France, UK, and the US to founding Nova and co-founding Aethos, a nonprofit AI innovation hub fostering collaboration between researchers, entrepreneurs, and artists.

TOPICS DISCUSSED:

1. Jean's Unique Background
From an academic background in philosophy to rock band musician to AI founder, Jean explains how his versatile education and ADHD contributed to his multidisciplinary approach to innovation.

2. The Birth of Nova
Jean shares how his transformative experience at Stanford led him to create Nova, an AI-powered research tool that helps researchers navigate scientific literature more efficiently and combat misinformation in academic papers.

3. Aethos: A Community for Responsible AI
How a nonprofit AI innovation hub came to life, bringing together founders, researchers, and artists committed to building human-centered AI solutions across multiple locations, starting in Cambridge.

4. Ethics in AI Development
Jean discusses his approach to evaluating startups for ethical considerations, the importance of transparency in AI model training, and how to implement responsible practices in AI development.

5. Digital Renaissance
The philosophical concept that AI can augment human capabilities and help us become "multi-experts" like Renaissance figures, enabling a new era of human flourishing if anchored in humanistic values.

6. Copyright and Intellectual Property
Jean shares his contrarian view that intellectual property ultimately belongs to humanity, challenging conventional notions of copyright and ownership in the AI age.

7. Community-Driven Innovation
How Aethos fosters peer-to-peer learning, self-organization, and collective intelligence through initiatives like "pods" and "Unconferences."

INSIGHTS:

- The integration of philosophy, art, and technology creates a more holistic approach to AI development
- Transparency in AI training is crucial for building responsible AI systems
- Peer-to-peer learning and community intelligence often yields better results than traditional top-down leadership
- AI can help us become "multi-experts" and achieve greater human flourishing
- "There is no point to have a technology without consciousness" (adapting Montaigne's philosophy)
- The artist/founder ego is often overrated; creation should benefit humanity as a whole

CONTACT INFO:

- LinkedIn: Jean Arnaud
- Organizations: Aethos, Nova

CHAPTERS
02:01 From Philosophy Professor to AI Founder
05:00 The Three Pillars: Education, Art, and Entrepreneurship
08:54 The Birth of Nova
10:13 Nova's Evolution and Pivot
15:14 Founding Aethos as a Nonprofit AI Hub
17:04 The Mission of Responsible AI Innovation
19:56 Evaluating Ethical AI Startups
22:51 Implementing Responsible AI in Practice
26:45 Transparency in AI Model Training
29:03 Rethinking Copyright in the AI Age
32:25 The Future of Ownership and Decentralization
35:58 Creating a Collaborative Innovation Environment
40:35 The Power of Community Intelligence
43:43 Building Your Own Meaning Through Impact
49:37 The Concept of Digital Renaissance

Follow us at AI Ketchup for bi-weekly stories of AI builders and founders turning ideas into successful tech products.

Join us for an insightful conversation with Kavita Ganesan, an experienced AI strategist who has built solutions for Fortune 500 companies like 3M and eBay. Kavita shares her journey in AI and provides practical frameworks for organizations looking to implement AI successfully, including her B-CIDS framework (Budget, Culture, Infrastructure, Data, and Skills) and guidance on evaluating AI pilots.

TOPICS DISCUSSED:

1. Kavita's AI Journey
From academic research at USC to practical implementations at eBay and 3M, Kavita shares how she developed her unique perspective as a "translator" between business and technical worlds.

2. The B-CIDS Framework
A comprehensive approach to AI readiness focusing on Budget, Culture, Infrastructure, Data, and Skills, with special emphasis on data and cultural readiness as foundational elements.

3. Data Readiness Challenges
The critical importance of digitizing paper processes, comprehensive data collection, and unified data warehousing across company branches.

4. Cultural Readiness and AI Literacy
Balancing enthusiasm and fear through company-wide AI literacy programs to enable better collaboration and understanding of AI risks.

5. Problem-First Approach to AI
Why business leaders should focus on identifying real business problems rather than forcing AI adoption without clear use cases.

6. AI Pilot Success Metrics
The three pillars of successful AI implementation: model performance, business outcomes, and user experience.

7. Recommended AI Use Cases
Sector-specific recommendations such as recommendation systems for e-commerce and content creation tools for marketing teams.

INSIGHTS:

- Data readiness and cultural readiness take the longest to implement and should be prioritized
- AI solutions should be built with production constraints in mind from the beginning
- Companies should avoid hiring data scientists without clear business problems to solve
- The costs and risks of third-party APIs need careful consideration in pilot projects
- Traditional machine learning tools are often more predictable and easier to implement than generative AI
- Business leaders should focus on problems first, then determine if AI is the appropriate solution

CONTACT INFO:

- Book: The Business Case for AI
- Website: kavita-ganesan.com

CHAPTERS
00:00 Introduction to data readiness challenges
00:59 Welcome and guest introduction
01:39 Kavita's background
02:25 Kavita's journey in AI
05:11 Introduction to the B-CIDS framework
05:54 Applying B-CIDS to mid-sized companies
09:18 The unique challenge of cultural readiness
10:41 Focus on business problems, not just AI
12:29 Common pitfall: hiring data scientists without clear problems
14:53 Why AI pilots fail
17:08 Three pillars of AI success evaluation
20:50 Recommended AI use cases
22:01 The value of different AI tools beyond ChatGPT
23:49 Ethical concerns and risks
25:40 Finding balance between innovation and risk
26:22 Closing thoughts

Follow us at AI Ketchup for bi-weekly stories of AI builders and founders turning ideas into successful tech products.

Join us for an insightful conversation with Sebastian Raschka, a renowned machine learning expert and author who has significantly contributed to AI education through his book "Build a Large Language Model from Scratch." Sebastian shares his journey in machine learning, offers advice for newcomers to the field, discusses the latest advancements in reasoning models, and explores the future of model architectures.

TOPICS DISCUSSED:

1. Learning AI from ScratchSebastian discusses effective approaches to learning AI today, emphasizing the importance of finding balance between theory and practical projects, and maintaining focus despite the overwhelming amount of available resources.

2. Reading Scientific PapersInsights on how Sebastian approaches scientific literature, his method for filtering relevant papers, and how he extracts valuable information without getting lost in the flood of new research.

3. Reasoning ModelsAn exploration of reasoning models, their practical applications, and how they differ from traditional LLMs in providing step-by-step solutions for complex problems.

4. Future of Model ArchitecturesSebastian discusses the evolution of transformer architectures, state space models like Mamba, and Google's Titan models, offering his perspective on where architectural innovation is heading.

5. Multi-GPU Training EnvironmentsPractical insights into the challenges of training large models on multiple GPUs, including hardware considerations and the realities of resource-constrained environments.

6. Open-Source ContributionsSebastian shares his experience working with PyTorch founders at Lightning AI and discusses how open-source projects can be sustainable while balancing commercial interests.

INSIGHTS:- Find a project that excites you to stay motivated when learning AI and balance learning theory with practical application- Reasoning models excel at tasks requiring step-by-step solutions, particularly for code and math problems- The ability to toggle reasoning capabilities on and off is becoming a standard feature in modern LLMs- The pre-training paradigm may be reaching saturation, with more opportunities in post-training approaches- Open-source contributions create synergies that benefit both companies and the broader community

FURTHER POINTERS:- Article on Reasoning Models: State of LLM Reasoning and Inference Scaling- Sebastian's Book: Build a Large Language Model from Scratch- Lightning AI platform

CONTACT INFO:- GitHub: Sebastian Raschka- LinkedIn: Sebastian Raschka

CHAPTERS
00:46 Introduction to Sebastian's career
02:27 Learning AI from scratch in 2025
07:47 Managing information overload and learning resources
10:48 Approaching scientific papers effectively
14:02 Reading papers with a purpose
17:38 Reasoning models and their applications
27:26 Future of LLM integration in applications
29:35 Future of model architectures beyond transformers
37:36 Evolution of pre-training and post-training approaches
40:18 Multi-GPU environments and challenges
48:44 Balancing open source with commercial interests
55:24 Closing recommendations

Follow us at AI Ketchup for bi-weekly stories of AI builders and founders turning ideas into successful tech products.