In this episode, you’ll explore the field of deep reinforcement learning and the ways it influences the real world with Anna Goldie, Senior Staff Research Scientist at Google DeepMind. 

Anna’s current role has her working on Large Language Model (LLM) research for Gemini and Bard. Prior to that, she worked on reinforcement learning for LLMs and retrieval-augmented LLMs at Anthropic, and was co-founder/lead of the ML for Systems team in Google Brain.

During this wide-ranging discussion, you’ll learn about her contributions to the field of reinforcement learning, and how we can leverage reinforcement learning effectively for real world applications going forward. 

Sponsored by: https://odsc.com/ 
Find more ODSC lightning interviews, webinars, live trainings, certifications, bootcamps here – https://aiplus.training/ 

Topics:
1. Professional journey and the key moments 
2. Core principles of deep reinforcement learning
3. Deep reinforcement learning for chip design vs  traditional approaches
4. Key complexities in modern chip design and how deep reinforcement learning can address these complexities
5. Discuss Google’s TPUs - Tensor Processing Units - built specifically for accelerating machine learning workloads. 
6. The potential of Deep Reinforcement learning in computer systems or other domains within Google Deepmind.
7. Deep reinforcement learning use in Large Language Models (LLMs)
8. Reinforcement Learning from Human Feedback (RLHF), designing effective rewards and providing feedback at scale
9. Scalable supervision techniques, for developing methods to efficiently gather feedback that aligns the LLM with human preferences
10. Implement the Constitutional AI framework where AI models are guided by a set of foundational principles or 'constitutional' directives
11. How Retrieval Augment Generation (RAG) systems improve the accuracy and relevance of responses compared to standard large language models even LLMs with large retrieval context windows
12. How “RAPTOR: RECURSIVE ABSTRACTIVE PROCESSING FOR TREE-ORGANIZED RETRIEVALcompares to traditional RAG approaches that retrieve short, contiguous chunks?
13. Hierarchical summaries with RAPTOR
14. LLM Finetuning With Low-Rank Adaptation (LoRA)
15. Google's Gemini 1.5 next-generation LLM model and mixture of experts architecture
16. CALM—Composition to Augment Language Models
17. How dual undergraduate degrees in both computer science and linguistics from MIT has contributed to your innovative work in machine learning,
18. Constitutional AI at Antropic https://www.anthropic.com/news/claudes-constitution
19. What is the best way to follow your work?
20. Keynote address at ODSC East in mid-April. 


SHOW NOTES
More about Anna Godie
https://www.linkedin.com/in/adgoldie/
https://www.annagoldie.com/


More about Constitutional AI at Antropic
https://www.anthropic.com/news/claudes-constitution
Constitutional AI: Harmlessness from AI Feedback
 https://arxiv.org/pdf/2212.08073.pdf


More about Anna’s Paper
RAPTOR: RECURSIVE ABSTRACTIVE PROCESSING FOR TREE-ORGANIZED RETRIEVAL
https://openreview.net/pdf?id=GN921JHCRw
The official Code implementation of RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval
https://github.com/parthsarthi03/raptor


More about Large Language Models
Direct Preference Optimization: Your Language Model is Secretly a Reward Model
https://arxiv.org/abs/2305.18290
LLM Finetuning With Low-Rank Adaptation (LoRA)
https://lightning.ai/pages/community/article/lora-llm


CALM—Composition to Augment Language Models
https://arxiv.org/pdf/2401.02412.pdf
https://www.anthropic.com/news/claudes-constitution

Deep Reinforcement Learning in the Real World with Anna Goldie

In this episode, we speak with Brooke Hopkins, Founder and CEO of Coval, to unpack why Voice AI reliability is starting to look a lot like the reliability problem self-driving teams have been solving for years. Brooke shares how her work at Waymo building dataset and simulation tooling shaped Coval’s approach to testing voice agents, and why voice systems fundamentally change the ML/data science playbook—breaking i.i.d. assumptions, making labels more subjective, and pushing teams toward trajectory-based analysis, simulation-driven testing, and continuous production monitoring.
Key Topics Covered:
Why Voice AI and self-driving share the same core challenge of building trust in non-deterministic systems
Moving from autonomy to Voice AI and why timing (models + latency) finally made voice viable
What to simulate for Voice AI: accents, languages, background noise, interruptions, and workflow variability
How to turn a small test set into thousands of scenarios using personas, transcripts, and configurable metrics
Failure modes in Voice AI: compounding errors, misclassified intent, looping, abandonment, and “looks fine but feels bad” experiences
Why conversations violate i.i.d. and how that forces teams to evaluate multi-turn trajectories, not single responses
Building a pragmatic quality system: simulate pre-production and monitor in production 
How to think about “minimum viable” testing vs. mature regression suites 
What’s next: faster, more targeted voice architectures, multi-agent systems, and new engineering roles around real-time voice
Memorable Outtakes:
Brooke: “In the same way that every company has a website, every company is expected to have a voice experience.” 
Brooke: “How do you create reliable voice agents when everything is sand?”
Brooke: “How do you get to that really quality voice agent, especially once you get past that, okay, it works, it's low enough latency, et cetera, but how do we get to this long tail of issues?”


References & Resources:
Brooke Hopkins (Coval) – https://www.linkedin.com/in/bnhop/
Coval – https://coval.dev
Coval (Y Combinator): https://www.ycombinator.com/companies/coval
Voice Activity Detection chapter (speech processing book): https://speechprocessingbook.aalto.fi/Recognition/Voice_activity_detection.html
i.i.d. (Independent and identically distributed random variables): https://en.wikipedia.org/wiki/Independent_and_identically_distributed_random_variables
Regression testing: https://en.wikipedia.org/wiki/Regression_testing
Computer simulation: https://en.wikipedia.org/wiki/Computer_simulation
Dialogue system / conversational agent: https://en.wikipedia.org/wiki/Dialogue_system
Speech recognition (ASR/STT): https://en.wikipedia.org/wiki/Speech_recognition
Text-to-speech (TTS): https://simple.wikipedia.org/wiki/Text_to_Speech
Sponsored by:
This episode was sponsored by:
ODSC AI East 2026 – The Leading AI Training Conference
Join us in Boston from April 28th–30th for expert-led sessions on Agentic AI, AI Engineering, Data Science, Machine Learning, LLMOps, and AI-driven automation.
Learn more: https://odsc.ai/east
Agentic AI Summit – Learn to Build and Deploy Autonomous AI Systems
Happening online this January, the Agentic AI Summit is your chance to dive into hands-on workshops and expert sessions on AI agents, workflows, and real-world automation.
Use the code podcast for 10% off any ticket for either event
Learn more: https://summit.ai



Building Trustworthy Voice AI: The Evaluation Playbook with Brooke Hopkins

In this episode of the ODSC AiX Podcast, host Sheamus McGovern sits down with Sinan Ozdemir — AI author, consultant, entrepreneur, and lecturer — to cut through the hype around AI agents and explore what truly works in real‑world AI deployments. With over a decade of experience advising companies on AI strategy and building production AI systems, Sinan offers grounded insights from his book Building Agentic AI Workflows. They discuss the practical differences between workflows and agents, why agents aren’t always the right choice, how fine‑tuning can improve performance, and how careful evaluation, tool design, and calibration can make AI systems more reliable and trustworthy.
Key Topics Covered:
• Why 2025 didn’t turn out to be the year of agents in the sense people hoped
• Defining workflows, agents, and “agentic workflows”
• Comparing agents and workflows on accuracy, latency, and cost
• The importance of well‑designed tools for agent success
• Understanding confidence and calibration in AI systems
• The role and limitations of MCP (Model Capability Protocol)
• Experiences and advice on choosing open source models for workflows/agents
• When to fine‑tune models versus relying on retrieval tools
• Best practices in fine‑tuning, including avoiding catastrophic forgetting
• Advice for experimenting with fine‑tuning on realistic budgets
• Practical considerations for building domain‑specific AI workflows
Memorable Outtakes:
• “Agents aren’t a magic replacement for workflows — they’re a tool, and often a workflow can do as well or better for less cost and latency.” — Sinan Ozdemir
• “When an AI says ‘I don’t know’ — that’s one of the best signs of a well‑calibrated system.” — Sinan Ozdemir
References & Resources:
Guest:
Sinan Ozdemir — AI author, consultant, lecturer, and AI workflow expert.
LinkedIn: https://www.linkedin.com/in/sinan-ozdemir
GitHub: https://github.com/sinanuozdemir
Personal website (workshops & training): https://sinanozdemir.ai/workshops
Book Mentioned:
Building Agentic AI Workflows — Pearson AI Signature Series (Sinan Ozdemir)
Tools & Frameworks Mentioned:
LangGraph (AI workflow and agent framework) https://www.langchain.com/langgraph
LangFlow (Low/no‑code AI workflow builder) https://www.langflow.org
Uns­loth (Efficient fine‑tuning framework) https://unsloth.ai
This episode was sponsored by:

ODSC AI East 2026 – The Leading AI Builders Conference
Join us in Boston from April 28th–30th for expert‑led sessions on Agentic AI, AI Engineering, Data Science, Machine Learning, LLMOps, and AI‑driven automation.
Learn more: https://odsc.ai/east
Agentic AI Summit – Learn to Build and Deploy Autonomous AI Systems
Happening online this January, the Agentic AI Summit is your chance to dive into hands‑on workshops and expert sessions on AI agents, workflows, and real‑world automation.
Use the code podcast for 10% off any ticket for either event
Learn more: https://summit.ai



AI Agents vs. Workflows: Reality, Hype & What Actually Works with Sinan Ozdemir

In this episode of the ODSC AiX Podcast, host Cal Al-Dhubaib sits down with Jane Alexander, Chief Digital Information Officer at the Cleveland Museum of Art, to explore how one of the world’s leading museums has become a global innovator in digital transformation. Jane shares her journey integrating emerging technologies such as AI, augmented reality, computer vision, and open data into museum experiences that make art more accessible, interactive, and inclusive. From the groundbreaking launch of Gallery One to the museum’s pioneering open access initiative and the newest AI-powered accessibility tools, this conversation highlights how technology can deepen engagement with art while staying rooted in mission, ethics, and visitor experience.


Key Topics Covered:
• What makes the Cleveland Museum of Art one of the top museums in the country
• The origins and impact of Gallery One as a pioneering tech-enabled museum space
• How Art Lens Gallery gamifies learning and removes intimidation from the museum experience
• Cleveland Museum of Art’s leadership in open access and why it matters globally
• Access to 68,000 artworks, high-resolution images, metadata, and public APIs for free use
• The importance of strong data infrastructure, governance, and “one source of truth” systems
• Lessons in change management and bringing museum teams along in digital transformation
• Early uses of AI in museum interactives like Make a Face and Match a Pose
• AI experiments launched during the pandemic, including Art Lens for Slack and Share Your View
• Using AI ethically and transparently to enhance – not replace – the human experience of art
• The Renaissance to Runway exhibition video project and public misconceptions around AI art
• The museum’s next frontier: AI-powered accessibility, including visual descriptions for all artworks
• Upcoming immersive Art Lens innovations launching in 2026
Memorable Outtakes:
• “We never use tech for tech’s sake. It’s always about the art.” — Jane Alexander
• “We are not owners of these objects — we are caretakers. The objects are actually all of yours.” — Jane Alexander
• “If an AI tool solves a problem and creates access, then it’s meaningful innovation.” — Jane Alexander
References & Resources:
Guest: Jane Alexander — Chief Digital Information Officer, Cleveland Museum of Art
LinkedIn:⁠ https://www.linkedin.com/in/janealexander/⁠
Cleveland Museum of Art:⁠ https://www.clevelandart.org
⁠
Museum Open Access & Data Resources: Cleveland Museum of Art Open Access GitHub:⁠ https://github.com/ClevelandMuseumArt/openaccess⁠
Cleveland Museum of Art Open Access Info:⁠ https://www.clevelandart.org/articles/half-decade-open-access⁠
AI Tools Overview at CMA:⁠ https://www.clevelandart.org/ai-tools-overview
⁠
Renaissance to Runway Exhibition Videos:
Renaissance to Runway Video – Part 1:⁠ https://www.instagram.com/reel/DQ4-V-vEtB4/?utm_source=ig_web_copy_link⁠
Renaissance to Runway Video – Part 2:⁠ https://www.instagram.com/reel/DQ9QQVDgI_z/?utm_source=ig_web_copy_link
⁠
This episode was sponsored by:
ODSC AI East 2026 – The Leading AI Training Conference. Join us in Boston from April 28th–30th for expert-led sessions on Agentic AI, AI Engineering, Data Science, Machine Learning, LLMOps, and AI-driven automation. Learn more:⁠ https://odsc.ai/east⁠
Agentic AI Summit – Learn to Build and Deploy Autonomous AI SystemsHappening online this January, the Agentic AI Summit is your chance to dive into hands-on workshops and expert sessions on AI agents, workflows, and real-world automation. Use the code podcast for 10% off any ticket for either eventLearn more:⁠ https://summit.ai⁠

Integrating AI, AR, and Open Access at the Cleveland Museum of Art with Jane Alexander

In this episode of the ODSC AIx Podcast, Sheamus McGovern speaks with Jerry Liu , Co-Founder & CEO, LlamaIndex,  about the shift from early “RAG frameworks” to document AI workflows and the practical layer enterprises need to turn PDFs, Word docs, PowerPoints, and spreadsheets into reliable, structured inputs for agents and automation. Jerry explains why legacy OCR and “naive RAG” often fail (loss of structure, bundled packets, fragile templates), how VLM-based parsing changes the game, and what it takes to scale document workflows—from small file-system-based prototypes to production retrieval across millions of documents.

Key Topics Covered:
Jerry’s background: ML engineering and research experience (Robust Intelligence,
Why LlamaIndex says “RAG 1.0 is dead” 
LlamaIndex’s current mission: document AI workflows  
Where legacy OCR breaks: tables, charts, layout, “coordinate-based fragility,” 
VLM-powered document understanding
Document parsing outputs that matter downstream:
Why “naive RAG” struggles with documents
The evolution from search/chat to agentic workflows that “do work” (process automation, exception handling, human-in-the-loop)
Enterprise readiness and Trends: coding agents and computer-use agents;
Practical advice: scope deterministic vs assistant use cases

Memorable Outtakes:
Jerry: “It didn’t matter how good the LLM was because you just weren’t able to actually do anything with this data.”
Jerry: “If there weren’t regulations, bureaucracy… you could start automating a good 30 to 50% chunk of an entire human’s day.”
References & Resources:
Jerry Liu (Co-Founder & CEO, LlamaIndex)
X (Twitter): https://x.com/jerryjliu0
LinkedIn: https://www.linkedin.com/in/jerry-liu-64390071/
LlamaIndex
X (Twitter): https://x.com/llama_index
Llama Cloud signup: https://cloud.llamaindex.ai
LlamaIndex 10,000 free credits
https://login.llamaindex.ai/sign-up

Sponsored by:
This episode was sponsored by:
🔥 ODSC AI East 2026 – The Leading AI Training Conference
Join us in Boston from April 28th–30th for expert-led sessions on Agentic AI, AI Engineering, Data Science, Machine Learning, LLMOps, and AI-driven automation.
Learn more: https://odsc.ai/east


Document AI Workflows: Automating Real-World Document Work with Jerry Liu

AI has hit a real turning point: success isn’t just about model accuracy anymore, as it’s about how well the full system works together. In this episode, Sheamus McGovern sits down with Michelle Bonat (Head of AI at Golden Pear Funding; former Chief AI Officer & CTO at AI Squared) to explore what it takes to move from AI vision to orchestrated, observable, and governed systems that drive measurable business results. Drawing on experience leading large AI programs at firms like JPMorgan Chase and in venture-backed startups, Michelle shares practical frameworks for building AI that’s reliable, explainable, auditable, and goal-aligned—and highlights where the real innovation is happening today, from agent coordination and real-time monitoring to compliance automation and trust frameworks.
Key Topics Covered:
Why “models don’t create value.”
The shift from model race to the systems/orchestration era
What orchestration means in practice for real AI programs
The “system layer”: context, governance, observability, trust
The paradox of AI observability and what to measure in production
Governance as leverage (not overhead) for faster, safer deployment
Avoiding “POC purgatory” by tying AI to one owner and one business metric
Memorable Outtakes from Michelle Bonat :
“Models don’t create value. Systems do.”
“Trust in AI is not a feeling. It’s a behavior.”
“Autonomy without orchestration is just automated risk.”
“Orchestration is what turns AI from a science project into infrastructure.”
References & Resources:
Michelle Bonat (LinkedIn): https://www.linkedin.com/in/mbonat/
Michelle Bonat (Website): https://michellebonat.com/
Who is Watching Your AI Agents? How To Do AI Observability with AI (The Paradox of AI Observability): https://www.linkedin.com/pulse/who-watching-your-ai-agents-how-do-observability-michelle-bonat-zplyc/
A Step by Step Guide to Creating your Organization’s AI Playbook: Boat or Moat?: https://www.linkedin.com/pulse/step-guide-creating-your-organizations-ai-playbook-boat-bonat-vqjaf
Book reference: Agentic Artificial Intelligence: https://www.amazon.com/Agentic-Artificial-Intelligence-Harnessing-Reinvent-ebook/dp/B0F1DS36YC
Sponsored by:
This episode was sponsored by:
🔥 ODSC AI East 2026 – The Leading AI Training Conference
Join us in Boston from April 28th–30th for expert-led sessions on Agentic AI, AI Engineering, Data Science, Machine Learning, LLMOps, and AI-driven automation.
Use the code podcast for 10% off.
Learn more: https://odsc.ai/east



Orchestrating AI for Business Outcomes with Michelle Bonat

In this episode of the ODSC AI Podcast, host Cal Al-Dhubaib sits down with Salema Rice, a pioneering Chief Data & AI Officer and founder of Intelecon Global. With more than 22 years of executive experience leading data and AI at organizations including Allegis Group, Bain Capital, Accenture, and in partnership with Microsoft, Salema shares what’s really happening behind the scenes in enterprise AI transformation.
The conversation explores the human side of AI leadership in 2026—burnout at the executive level, the “frozen middle” in organizations, the importance of data foundations, and why compassion and culture are the true differentiators for long-term AI success. Salema also discusses her work building a global community of 500+ senior data, AI, and technology leaders through Intelecon Global and previews her upcoming data leadership initiative with Microsoft.
Key Topics Covered:
The real pressures facing Chief Data & AI Officers in 2026
Executive burnout and the growing expectations from boards and CEOs
The “frozen middle” problem in AI-driven transformation
Why AI success starts with responsible data and strong data governance
“Garbage in, garbage out” in the era of generative and agentic AI
The importance of building trusted data foundations before scaling AI
The origin and global expansion of Intelecon Global
Peer-to-peer executive problem-solving and experience-sharing forums
Leading with compassion in technical and executive roles
How culture remains the #1 challenge for Chief Data Officers
Hiring beyond tools: focusing on problem solvers over specific technologies
The shift from dashboards to real-time, action-oriented AI insights
Moving from AI hype and pilots to measurable business value
The evolving role of the Chief Data & AI Officer across industries
Why data and AI leaders must think like business value drivers, not cost centers
Talent strategy, upskilling, and the future of AI-augmented work
Memorable Outtakes:
On Data Foundations and AI
“AI didn’t abruptly start in November of 2022 with ChatGPT. We’ve been doing AI for a long time. The critical factor to responsible AI is responsible data.”
On Leadership and Culture
“If you want to trust the data, you have to first trust the data leader. And trusting the data starts with trusting me.”
On AI and Human Work
“Every job in the world will be X plus AI in the future. But AI can’t replace the human moments that matter most.”
On Mindset and Possibility
“There’s a lot of magic and pixie dust in the art of possible. Even if you can’t do it all today, it doesn’t mean you can’t do it.”
References & Resources:
Salema’s LinkedIn: https://www.linkedin.com/in/sjrice/
Cal’s LinkedIn: https://www.linkedin.com/in/dhubaib/
Intelecon Global (Official Website): https://intelliconglobal.com/
Sponsored by:
🔥 ODSC AI East 2026 – The Leading AI Training Conference
Join us in Boston from April 28th–30th for expert-led sessions on Agentic AI, AI Engineering, Data Science, Machine Learning, LLMOps, and AI-driven automation.
Learn more: https://odsc.ai/east
Use the code podcast for an additional 10% off.


What's on the Chief AI and Data Officer's Mind in 2026 with Salema Rice

In this episode, host Sheamus McGovern sits down with Joseph B. Fuller, Professor of Management Practice at Harvard Business School, founder and co-leader of HBS’s Managing the Future of Work initiative, to unpack a core paradox: why companies insist they can’t find talent while qualified workers struggle to get hired. Fuller explains how “spot-market, real-time hiring,” weak workforce planning, credential inflation, and broken screening systems created today’s hiring gridlock. We also discuss how AI (especially agentic AI and decision intelligence platforms) may accelerate task displacement, reshape talent pipelines, and force organizations to redesign work rather than “bolt on” tools.
Guest: Joseph B. Fuller (Harvard Business School Faculty Page): https://www.hbs.edu/faculty/Pages/profile.aspx?facId=123284
Key Topics Covered:
The “talent paradox”: why employers say they can’t hire while qualified workers struggle to get hired
How “spot-market” recruiting replaced workforce planning—and why it breaks
Credential inflation and why degree requirements became a default hiring proxy
Demographics, caregiving, and the shrinking labor supply across developed economies
Why “soft skills” drive churn, but most hiring processes don’t measure them well
How AI changes work at the task level, and why organizations must redesign workflows
Agentic AI and decision-intelligence layers: moving from data → decisions → actions faster
The risk to entry-level roles and the talent pipeline as AI boosts senior productivity
“Context engineering” as the real differentiator
Practical pathways to opportunity: skills-based hiring, apprenticeships, and competency-based education models
Joseph Fuller Memorable Outtakes:
On the real AI skill: "I'm urging my companies to stop calling people prompt engineers and to call them context engineers. 
On the two types of AI companies: "My shorthand is that the first pool of companies are AI efficiency companies, and the second pool are AI abundance companies. About 12% of jobs in the United States, AI is already more efficient than an entry-level worker. Both the abundance and the efficiency of the company will just take the win. But the very important distinction has to do with workforce demographics and the necessity of human interaction with AI."
On AI outcomes: "There's this huge focus on hallucination results, but if you look at the distribution of results, it's a bell curve. There are plenty of times when the AI is generating an insight that is genuinely valuable,  
References & Resources:
Joseph B. Fuller Faculty Page: https://www.hbs.edu/faculty/Pages/profile.aspx?facId=123284
Managing the Future of Work (Harvard Business School): https://www.hbs.edu/managing-the-future-of-work/Pages/default.aspx
About the Project (Managing the Future of Work): https://www.hbs.edu/managing-the-future-of-work/about-the-project/Pages/default.aspx
Project on Workforce at Harvard (home): https://pw.hks.harvard.edu/
No Country for Young Grads (Burning Glass Institute): https://www.burningglassinstitute.org/research/no-country-for-young-grads
The Expertise Upheaval—How Generative AI’s Impact on Learning Curves Will Reshape the Workplace (HBS item page): https://www.hbs.edu/faculty/Pages/item.aspx?num=68582
The Expertise Upheaval (Burning Glass Institute): https://www.burningglassinstitute.org/research/the-expertise-upheaval
Sponsored by:
This episode was sponsored by ODSC AI East 2026 – The Leading AI Training Conference
Join us in Boston from April 28th–30th for expert-led sessions on Agentic AI, AI Engineering, Data Science, Machine Learning, LLMOps, and AI-driven automation.
Learn more: https://odsc.ai/east 




The AI Hiring Gap: Skills, Signals, and “Hidden Workers” Joseph Fuller

In this episode of the ODSC AIx Podcast, host Sheamus McGovern sits down with David vonThenen, Senior AI/ML Engineer in the Office of the CTO at NetApp. David is a seasoned keynote speaker and open-source contributor with deep expertise in Agentic AI, deep learning, model optimization, cloud-native architectures, and retrieval-augmented generation. Drawing on a career that spans enterprise storage, Kubernetes, and conversational AI, David brings a uniquely grounded, hardware-aware perspective to the fast-moving world of AI systems engineering.

The conversation covers the real-world gap between AI demos and production deployments, the industry-wide pivot toward efficiency and smaller models, quantization fundamentals, Hybrid RAG architectures combining vector embeddings and knowledge graphs, multi-agent protocols including MCP and Agent-to-Agent, and the critical role of governance and observability in regulated industries.

Key Topics Covered
The demo-to-production gap: why RAG and agentic systems that shine in demos often break in real-world deployments
The efficiency paradigm shift: industry's move from parameter-count maximalism to smarter-per-watt models
The DeepSeek moment: how a cost-efficient open-weight model rattled markets and reframed AI infrastructure investment
Quantization explained: what happens when you move from FP16 to INT8 or INT4, including accuracy tradeoffs and hardware compatibility caveats
Fine-tune first, quantize second: David's preferred workflow for building specialized, cost-efficient agents
Small Language Models (SLMs) as preferred engines for multi-agent systems: cost, speed, reduced hallucination, and domain focus
Microservices for cognition: why SLM-powered agents mirror the design principles of microservice architecture
Agent swarms and ensemble voting: use cases where redundancy and majority-voting models add genuine value
Hybrid RAG: combining vector embeddings with knowledge graphs to add data relationships and improve reasoning quality
Is SaaS dead? How agentic AI is reshaping software consumption and the enduring role of human judgment

Memorable Outtakes
1. David on why SLMs reduce hallucination:
"That subject matter expert is geared towards answering one particular domain. It's much easier to get those small language models to say, 'I don't know what you're talking about.' And by doing that, you can improve the answers overall for your agentic systems when you string these things together."
2. David on the microservices analogy for multi-agent AI:
"We're talking about microservice architecture for language models. You're creating purpose-built microservices that handle one specific thing. That's the way I like to think about it."

Resources:
LinkedIn: https://www.linkedin.com/in/davidvonthenen/
GitHub: https://github.com/davidvonthenen
NetApp: https://www.netapp.com
bitsandbytes (quantization library by Hugging Face): https://github.com/bitsandbytes-foundation/bitsandbytes
Unsloth (open-source quantization and fine-tuning library): https://github.com/unslothai/unsloth
Past ODSC Workshops & Code Resources
Workshop: Adaptive RAG Systems with Knowledge Graphs - Building Reinforcement-Learning-Driven AI Applications (ODSC East 2025): https://github.com/davidvonthenen/2025-odsc-east-workshop
Workshop: Rethinking RAG - How MCP and Agent2Agent Will Transform the Future of Intelligent Search (ODSC West 2025): https://github.com/davidvonthenen/2025-odsc-west-workshop

From the Host
Sheamus is currently writing The AI Skill Flip, a forthcoming book exploring how AI is reshaping the skills workers need to stay relevant — and how individuals and organizations can get ahead of the shift. Stay tuned for updates.

Sponsored By
ODSC AI East 2026 - The Leading AI Training Conference
Join us in Boston from April 28th-30th for expert-led sessions on Agentic AI, AI Engineering, Data Science, Machine Learning, LLMOps, and AI-driven automation.
Learn more: https://odsc.ai/east

Smarter Per Watt with David vonThenen

In this episode of the ODSC Ai X Podcast, host Alex Landa speaks with Benjamin Rosman, Professor at the University of the Witwatersrand and Director of the Machine Intelligence and Neural Discovery (MIND) Institute. Rosman is also Director of the RAIL Lab, Co-Founder of Lelapa AI, and Co-Founder of the Deep Learning Indaba. The conversation explores the growth of the AI ecosystem across Africa, the importance of community-driven initiatives like Deep Learning Indaba, and how researchers are tackling challenges such as low-resource languages, healthcare access, and agricultural innovation. Rosman also discusses interdisciplinary research at the MIND Institute and shares his vision for the next decade of AI development on the African continent.
Key Topics Covered:
- Benjamin Rosman’s journey into artificial intelligence and his early interest in computer science and gaming
- The evolution of AI research and education opportunities in Africa over the past two decades
- The founding and growth of Deep Learning Indaba and its role in building a pan-African AI community
- How grassroots initiatives and community networks are strengthening AI talent across the continent
- Real-world AI applications emerging from African researchers, including:
+ Healthcare diagnostics and medical prioritization systems
+ Agricultural disease detection using mobile devices
+ Financial inclusion and access to services
+ Educational tools for underserved communities
- The concept of frugal innovation and designing AI solutions that run on limited infrastructure or mobile devices
- Research challenges in Africa, including funding limitations, compute access, and infrastructure constraints
- The role of interdisciplinary collaboration at the MIND Institute
- AI research topics, ranging from cultural value alignment in AI systems to animal cognition and elephant communication
- The importance of social impact, policy engagement, and responsible technology deployment
- A vision for the future of Africa’s AI ecosystem and how young technologists can participate
Memorable Outtakes from Benjamin Rosman:
“There’s always do the coolest things you possibly can—ideally with the coolest people you can.”
“You go into places people might think of as completely backward or disconnected, and you find people doing technical work I couldn’t even imagine doing.”
“If you can just give young people the tools and resources, they’ll figure out the cool things to build.”
Resources:
1. LinkedIn: https://www.linkedin.com/in/benjamin-rosman-a734792b/
2. Machine Intelligence and Neural Discovery (MIND) Institute: https://www.wits.ac.za/mind/
3. The Deep Learning Indaba: https://deeplearningindaba.com/
4. Robotics, Autonomous Intelligence and Learning (RAIL) Lab: https://www.raillab.org/
5. Lelapa AI: https://lelapa.ai/
6. TIME Magazine Feature: https://time.com/collections/time100-ai-2025/7305865/benjamin-rosman/ 
7. University of the Witwatersrand: https://www.wits.ac.za/

Sponsored by:
🔥 ODSC AI East 2026 – The Leading AI Training Conference
Join us in Boston from April 28th–30th for expert-led sessions on Agentic AI, AI Engineering, Data Science, Machine Learning, LLMOps, and AI-driven automation.
Learn more: https://odsc.ai/east


Building the AI Ecosystem in Africa: Innovation, Community, and the Future of Intelligence with Benjamin Rosman

In this episode, Sheamus McGovern speaks with Robert Brennan, Co-Founder and CEO of OpenHands, about the rapid evolution of AI coding agents and how they are reshaping software development. Robert shares insights on how developer workflows are changing, from single-agent pair programming to large-scale multi-agent systems operating in parallel. The conversation explores the rise of open source AI, model-agnostic development, emerging roles like “agent orchestrators,” and the practical realities of deploying AI safely in enterprise environments.
Speaker:
Robert Brennan
GitHub: https://github.com/OpenHands
Website: https://www.openhands.dev
Twitter/X: https://twitter.com/brandondbrennan
Key Topics Covered:
How AI coding agents have improved with newer models and become part of daily development workflows
The shift from developers writing all code to supervising and guiding AI-generated code
Differences in enterprise adoption of AI—forward-thinking vs. cautious organizations
Why multi-agent “teams” haven’t worked as expected—and the rise of subagents and parallelization
Using parallel agents to dramatically speed up tasks like vulnerability remediation
The emergence of new roles such as “agent orchestrators” and “agent pilots”
Local (IDE-based) vs. cloud-based agent workflows and tooling differences
The importance of observability and measuring agent performance at scale
Why open source and model-agnostic approaches are becoming more important
The OpenHands Index and challenges with benchmarking AI coding performance
Risks and realities of open source models, including data sovereignty and geopolitical considerations
The concept of inner loop vs. outer loop development and where agents provide the most leverage
Real productivity gains (20–40%) vs. hype around AI replacing developers
Security risks and lessons from experimental open agent systems like OpenClaw
Practical advice for engineers: experiment, build custom workflows, and stay hands-on
Memorable Outtakes:
“These agents are just getting more and more useful,and harder to ignore in terms of the value they create.”
“We’re used to writing every line of code ourselves—and that is changing. That’s a little scary.”
“If you’re trying to adopt AI at scale, you should be trying to get more done—not just cut your workforce.”
References & Resources:
OpenHands: https://www.openhands.dev
OpenHands GitHub: https://github.com/OpenHands
Laminar (AI observability): https://github.com/lmnr-ai/lmnr
OpenHands Index (AI model benchmarking): https://www.openhands.dev/index
Agent Skills standard: https://agent-skills.ai
ARC Prize: https://arcprize.org
Sponsored by:
This episode was sponsored by:
🔥 ODSC AI East 2026 – The Leading AI Training Conference
Join us in Boston from April 28th–30th for expert-led sessions on Agentic AI, AI Engineering, Data Science, Machine Learning, LLMOps, and AI-driven automation.
Learn more: https://odsc.ai/east

Open Source Coding Agents: A New Developer Workflow with Robert Brennan

With Ai X Podcast, Open Data Science Conference (ODSC) brings its vast experience in building community and its knowledge of the data science and AI fields to the podcast platform. The interests and challenges of the data science community are wide ranging. To reflect this Ai X Podcast will offer a similarly wide range of content, from one-on-one interviews with leading experts, to career talks, to educational interviews, to profiles of AI Startup Founders. Join us every two weeks to discover what’s going on in the data science community.

Find more ODSC lightning interviews, webinars, live trainings, certifications, bootcamps here - https://aiplus.training/  
Don't miss out on this exciting opportunity to expand your knowledge and stay ahead of the curve.

Deep Reinforcement Learning in the Real World with Anna Goldie

ODSC's Ai X Podcast