In this episode, we talk with Bassam Tabbara, Founder and CEO at Upbound, Founder of Crossplane, and Founder of Modelplane, about what it takes to run AI inference at scale.

Bassam walks us through his path from writing BASIC on a ZX Spectrum in Lebanon to building the early automation behind Hotmail at Microsoft, founding Crossplane, and then Modelplane: a new open-source fleet orchestrator built for running any model, on any engine, on any hardware.

KEY TAKEAWAYS

00:00 Alex introduces Bassam Tabbara, inventor of Rook and Creator of Crossplane, here to discuss his newest release, Modelplane.

01:20 Bassam's first exposure to programming: teaching himself BASIC on a ZX Spectrum 48K as a kid in Lebanon.

05:35 Control planes exist because human-in-the-loop failure response doesn't scale.

07:57 Owning your own intelligence is fundamentally about data control, not just cost.

11:05 Modelplane was built because Upbound's enterprise customers are already running inference across multiple Kubernetes clusters.

13:45 The AI stack — models, serving engines, infrastructure, and accelerators — is tipping horizontal, with more vendor choice at every layer.

22:35 The fleet-wide inference problem was unsolved in open source, so Upbound built Modelplane to close that gap.

28:45 Modelplane separates concerns between platform teams and ML teams through two distinct APIs.

39:35 AI is a workload, not a new platform, and that reframe changes how enterprises architect around it.

48:25 Modelplane is an early, 0.1 release, and Bassam invites the open-source community to contribute.

Thanks for listening to "Hello Agent!: The podcast at the intersection of data & agents." If you loved this episode, let us know with a 5-star review! Remember to subscribe so you don't miss an episode. To learn more about Redpanda, visit redpanda.com

RESOURCES MENTIONED

ModelPlane, Bassam's newly released fleet orchestrator for AI inference:

modelplane.ai

Crossplane, the open-source control plane framework Bassam created:

https://www.crossplane.io

Rook, Bassam's earlier open-source storage project:

https://rook.io/

Upbound website:

https://www.upbound.io/

Cloud Native Computing Foundation (CNCF), which now hosts Crossplane as a graduated project:

https://www.cncf.io/

vLLM, one of the open-source serving engines discussed:

https://vllm.ai/

SGLang, the other serving engine discussed:

https://www.sglang.io/

#RealTimeData #DataStreaming #Redpanda

Most ML training bugs don't announce themselves. They hide in floating-point rounding, batch distribution drift, and concurrency issues that only emerge across thousands of GPUs. The hard part isn't writing the algorithm — it's building infrastructure disciplined enough to tell the difference between a numerically equivalent optimization and a quietly corrupted model.

We speak with Tianshu Yu, member of the technical staff at Liquid AI, about the real work of building reliable ML systems at scale. Tianshu previously led reinforcement learning infrastructure at ByteDance, where he was a core contributor to VERL — one of the most widely used open-source RL training frameworks. He now works on pre-training and architecture design for edge-deployed vision language models.

The conversation covers the three-phase structure of reinforcement learning, why checkpoints serve debugging, fault recovery, and model selection simultaneously, and how capability collapse emerges when data distributions concentrate during fine-tuning. We also get into what changes when you design models for edge hardware: the shift from attention to convolution-dominant ops, latency budgets as a first-class architectural constraint, and why a 350M parameter model has meaningfully different engineering requirements than a frontier model.

KEY TAKEAWAYS

00:00 Introduction.

03:10 Reinforcement learning trains models against reward signals rather than ground truth, requiring the system to generate its own training data through rollout.

06:00 Capability collapse is a real risk during RL fine-tuning: optimizing on a concentrated data distribution can degrade unrelated model capabilities, including multilingual performance.

09:00 Checkpoints serve three distinct purposes in large-scale training: monitoring quality curves, enabling fault recovery after hardware or software failures, and supporting model selection across training steps.

13:30 A smooth reward curve is not sufficient evidence of a healthy training run — benchmark evaluations at checkpoint boundaries frequently reveal instabilities invisible in aggregate metrics.

17:30 Training system bugs decompose into three categories: data pipeline issues, infrastructure logic errors, and algorithmic design flaws — and disentangling them requires iterative hypothesis testing, not a single diagnostic pass.

20:30 ML infrastructure correctness is fundamentally different from software correctness: because everything operates in floating point, a mathematically equivalent optimization can still produce divergent numerics.

24:00 Eval benchmarks are a late and noisy signal — 30-minute evaluation cycles and LLM-judge variance make them unsuitable as a primary debugging instrument during active training.

32:00 Open-source training frameworks like VERL provide two compounding benefits: a battle-tested abstraction layer that separates algorithm from infrastructure, and community validation that reduces the prior probability of fundamental kernel bugs.

39:00 Multimodal architectures unify vision and text by mapping continuous image patches into the same embedding space as discrete text tokens, enabling a shared transformer backbone across modalities.

45:00 Edge model architecture is constrained not just by parameter count but by hardware topology: devices without large tensor cores favor convolution-dominant operations over attention, requiring architecture decisions to be made alongside inference profiling.

Thanks for listening to “Hello Agent!: The podcast at the intersection of data & agents.” If you loved this episode, let us know with a 5-star review! Remember to subscribe so you don’t miss an episode. To learn more about Redpanda, visit redpanda.com

RESOURCES MENTIONED

VERL

github.com/verl-project/verl

Robust LLM Training Infrastructure at ByteDance

arxiv.org/abs/2509.16293

#RealTimeData #DataStreaming #Redpanda

Durable execution has been solving reliability problems for decades. Now it's the missing primitive for agentic AI systems.

We're joined by Jeremy Edberg, C-Suite Advisor* at DBOS, Inc., to explore how durable execution works, why it matters for building production-ready AI agents, and what lessons from Netflix-scale reliability engineering apply directly to the age of agentic coding.



*Since recording, Jeremy has become C-Suite Advisor to DBOS, Inc.

KEY TAKEAWAYS

00:00 Introduction 

01:30 DBOS was born from research into replacing the operating system with a database. 

04:00 A career spent making things reliable — from Reddit to Netflix — built Jeremy's obsession with automation. 

09:00 If you do something twice, automate it; make manual steps look like automation. 

12:30 Durable execution lets systems resume from exactly where they last succeeded. 

14:00 Replaying saved LLM outputs removes non-determinism from agentic debugging. 

16:00 In agentic coding, the bottleneck isn't writing code — it's deployment, testing, and operations. 

20:00 DBOS adds roughly 1% overhead to Postgres, making durable execution practical at scale.

28:00 Agentic AI created a new class of workloads between instant API calls and long bat

RESOURCES MENTIONED

DBOS, Inc

https://www.dbos.dev

Transact library

https://www.dbos.dev/dbos-transact

CockroachDB

https://www.cockroachlabs.com

Dosu

https://dosu.dev/

Thanks for listening to “Hello Agent!: The podcast at the intersection of data & agents.” If you loved this episode, let us know with a 5-star review! Remember to subscribe so you don’t miss an episode. To learn more about Redpanda, visit redpanda.com

#RealTimeData #DataStreaming #Redpanda

Dominik Tornow, Founder and CEO at Resonate HQ, discusses simplifying distributed systems and why durable execution is becoming critical in the age of AI agents.

We explore how agents introduce new challenges for distributed systems, why long-running autonomous workflows require interruption-tolerant execution, and how durable promises provide a simpler mental model for building reliable multi-agent systems.

Key Takeaways:

00:00 Introduction.

02:20 A career start during the rise of AWS and cloud computing shaped Dominik’s focus on distributed systems.

04:10 Separating the concept of durable execution from specific implementations led to the idea of interruption-tolerant systems.

05:15 Failures as interruptions — voluntary or involuntary — lead to a broader, more practical model for resilient execution.

12:40 Simplifying complex concepts and why engineers often struggle to align their mental models when designing distributed systems.

14:10 Dominik defines an agent as a combination of a model, tools, and a system prompt, while distinguishing between an agent definition and a running agent instance.

23:30 Many effective agents are simply tool-calling loops, despite the complexity of the surrounding ecosystem.

26:20 Running agents in enterprise environments introduces distributed systems challenges, especially when workflows involve long-running or expensive operations.

33:15 Unlike traditional microservices, retrying agent operations can be extremely expensive when LLM calls or data pipelines are involved.

40:20 A durable promise is a simple mechanism for coordination and recovery in distributed agent workflows.

Resources Mentioned:

Dominik Tornow

https://www.linkedin.com/in/dtornow/

Resonate HQ | LinkedIn

http://www.linkedin.com/company/resonatehqio/

Resonate HQ | Website

https://www.resonatehq.io/

"Think Distributed Systems" by Dominik Tornow

https://www.manning.com/books/think-distributed-systems

Temporal (durable execution platform)

https://temporal.io

Thanks for listening to “Hello Agent!: The podcast at the intersection of data & agents.” If you loved this episode, let us know with a 5-star review! Remember to subscribe so you don’t miss an episode. To learn more about Redpanda, visit redpanda.com

#RealTimeData #DataStreaming #Redpanda

Building reliable agents begins with a strong data foundation. As AI systems move from experiments to production, teams need clearer ways to connect data, models, and real-world decision-making. In this inaugural episode of “Hello, Agent!” we talk with Nicolas Dupont, Founder and CEO at Cyborg Inc., about how agents are changing the way teams think about automation, streaming data, and system design. Nicolas shares practical lessons from building Cyborg, what breaks when agents scale, and how engineers can design systems that stay flexible as complexity grows.

Key Takeaways:

00:00 Introduction.

02:00 Nicolas’ focus on solving technical problems for the real world.

05:43 Programming in C builds a deep appreciation for low-level systems.

09:39 Securing proprietary data matters more than models in enterprise AI.

15:32 Vector data can expose sensitive information to attackers.

20:59 Auth systems are often weaker than expected, even for sensitive data.

25:03 Agentic AI removes the perimeter, making zero trust essential.

30:06 Poisoned data is hard to trace across agent systems.

35:07 Deleting keys makes sensitive data unreadable immediately.

40:48 Modern cryptography is no longer a bottleneck, making strong encryption practical.

45:06 Zero trust isn’t optional — the future of agentic AI depends on it.

Resources Mentioned:

Nicolas Dupont

https://www.linkedin.com/in/cyborgnico/

Cyborg Inc. | LinkedIn

https://www.linkedin.com/company/cyborginc/

Cyborg Inc. | Website

https://www.cyborg.co/

Cyborg Blog: Vector Embeddings Are Not One-Way Hashes 

https://www.cyborg.co/blog/vector-embeddings-are-not-one-way-hashes

Thanks for listening to “Hello Agent!: The podcast at the intersection of data & agents.” If you loved this episode, let us know with a 5-star review! Remember to subscribe so you don’t miss an episode. To learn more about Redpanda, visit redpanda.com

#RealTimeData #DataStreaming #Redpanda