DataTalks.Club Podcast Wiki

Wiki

AI

How DataTalks.Club podcast guests define AI across machine learning, generative AI, agents, production systems, evaluation, infrastructure, and governance.

Related Wiki Pages

Generative AI LLMs AI Engineering AI Engineer Role Machine Learning Responsible AI and Governance

Definition

Artificial intelligence covers systems that automate or support decisions that would otherwise need human judgment. It includes recommendation, generation, search, and planning.

DataTalks.Club guests use AI for machine learning, deep learning, and NLP. Recent episodes also use AI for LLMs and generative AI. They also use AI for retrieval-augmented generation, AI agents, and production systems.

In Understanding the AI Engineer Role, Nasser Qadri separates the broad AI umbrella from the current industry habit of using AI to mean generative-AI products. He also ties current AI work to statistical evaluation and agents.

In From Game AI to LLM Agents, Micheal Lanham gives a longer lineage. Game AI and reinforcement learning sit inside the same historical family as modern LLM agents. So do evolutionary algorithms and NLP. Engineers build and evaluate each one differently.

Common Definition

Across recent episodes, guests treat AI as a system-level discipline rather than a single model call. A useful AI product combines model behavior with software, data, context, and evaluation. It also needs user experience, security, and operations.

In Paul Iusztin’s AI engineering episode, Paul Iusztin frames modern AI engineering as full-stack product work. The 22:29, 29:12, and 42:28 chapters put RAG and knowledge management in the same skill stack as agents. They also include LLMOps and product engineering. Shipping discipline sits alongside both.

Ruslan Shchuchkin makes the same point from a role perspective in Inside the AI Engineer Role. At 19:40, the AI engineer combines product discovery with backend work. The role also includes LLM tooling and delivery habits. That role framing links AI to AI Engineering, AI Engineer Role, and AI Engineering Roadmap.

The common definition is pragmatic. AI is useful when it changes a workflow or decision, not when a team merely adds a model endpoint. In Production AI Engineering, Bartosz Mikulski connects AI work to data trust, pipeline tests, and prompt evaluation from 9:05 through 28:16. The 30:00 and 31:45 chapters add prompt compression, caching, latency, and cost. That puts AI close to data engineering, MLOps, and production.

Guest Differences

Guests disagree less about whether AI matters and more about where the hard part sits. Nasser Qadri stresses statistical rigor and human-centered design in Understanding the AI Engineer Role, especially at 7:45 and 36:15. In his framing, teams need to ask what the system should optimize, how people use it, and how they know whether an answer is good.

Paul Iusztin puts more weight on the builder’s skill stack in his AI engineering episode. His 22:29 and 42:28 chapters cover full-stack engineering, product delivery, and technical pillars for shipping AI products. Ruslan Shchuchkin adds hiring and project evidence in Inside the AI Engineer Role. The 7:51 and 57:39 chapters make side projects and skill proof more important than credentials.

Andrey Cheptsov shifts the center toward infrastructure in Post-ChatGPT AI Infrastructure. The 8:25 and 21:37 chapters frame AI through cloud economics, privacy, and control. The 30:16 and 47:16 chapters add distributed training, GPU coordination, Kubernetes limits, and open-source orchestration. For infrastructure-heavy teams, the question isn’t only “which model?” It’s also who controls compute, who pays for it, and who operates it.

AI, Machine Learning, and Generative AI

Machine learning is a major technical subset of AI. In older DataTalks.Club episodes, AI often means predictive modeling, features, and labels. It also means training data, model evaluation and deployment. Use Machine Learning or Machine Learning Engineer Role when the discussion centers on supervised learning, model validation, feature pipelines, or classical ML system design.

Recent AI episodes often mean generative AI and LLMs. In Deploying LLMs in Production, Meryem Arik separates API-based LLM use and open-source models. She also covers fine-tuning, retrieval, vector databases, and deployment choices. Use that episode for LLM-specific design decisions. This topic is the broader AI map.

Generative AI covers systems that produce text, code, and images. It also covers summaries, answers, and plans.

Maria Sukhareva uses chatbot failures in Hardening Generative AI Chatbots to show why generated outputs need guardrails. The 9:28, 13:20, and 16:15 chapters cover prompt injection and data exfiltration. They also cover output validation and layered defenses. That makes generative AI part of security as well as product design.

Agents

Agents extend AI systems from answer generation into action. In Building Agentic AI Systems, Ranjitha Kulkarni defines agents through objectives, tools, and memory. She also covers knowledge stores, planning, and evaluation.

The 11:00 and 12:31 chapters cover objectives, tools, and memory. The 21:21 chapter adds context design. The 36:11 and 51:17 chapters connect agents to RAG, test datasets, and outcome-based checks.

Micheal Lanham adds a design history in From Game AI to LLM Agents. At 20:57 and 23:48, he discusses task decomposition, sequential flows, and manager-agent orchestration. At 31:31 and 57:39, he adds MCP-style tool integration and monitoring. Use Agent Engineering, AI Agents, and Multi-Agent Systems for the narrower designs.

Production Systems

Production AI work starts when the team asks how the system behaves after the demo. Bartosz Mikulski makes that practical in Production AI Engineering. Pipeline tests and prompt evaluation show up before the conversation reaches coding assistants. So do prompt compression, caching, and latency. Those production concerns connect AI to Production, LLM Evaluation Workflows, Data Quality and Observability, and Data Observability.

Retrieval changes production architecture, and Modern Search Systems shows why. In that episode, Atita Arora grounds RAG in search quality, embeddings, and chunking. She also covers citations and evaluation.

In Building Agentic AI Systems, Ranjitha Kulkarni warns at 29:30 and 31:38 that RAG inherits latency and cost problems. It also inherits data-quality problems. Use Search, Search and Knowledge Systems, Embeddings, and Vector Databases for those implementation details.

Infrastructure choices depend on cost, control, scale, and privacy. In Post-ChatGPT AI Infrastructure, Andrey Cheptsov compares cloud, on-prem, and open-source orchestration. He also covers distributed training and GPU coordination. That discussion anchors AI Infrastructure, Machine Learning Infrastructure, and MLOps.

Governance and Security

AI systems need governance when they affect users or regulated decisions. They also need governance when they use private data or operational workflows. In Responsible & Explainable AI, Supreet Kaur defines responsible AI through trust, fairness, and explainability. She also adds SME input and compliance.

The 27:38 and 35:28 chapters add monitoring and human oversight. These chapters connect AI governance to product choices, not only post-hoc documentation.

Security becomes sharper for LLM and chatbot systems. Maria Sukhareva shows in Hardening Generative AI Chatbots that prompt injection and knowledge-base leakage need specific defenses. Hallucinations and unsafe automation also need human review. That discussion belongs with Security, AI Red Teaming, Responsible AI and Governance, and Privacy Engineering for ML.

Use these pages for narrower AI topics and article-style guides.