Dominik Kundel

RAG is evolving from “AI answers with citations” into something more operational: Agentic RAG, where the system can orchestrate multi-step retrieval across sources and apply measurable checks like context relevance, answer relevance, and groundedness. This Q&A with Eudald Camprubí explains why that matters. Enterprise adoption hinges on trust, traceability, and governance as much as model quality. The payoff isn’t just Q&A. It’s powering practical workflows like content drafting that stays on-message, faster campaign and competitive research synthesis, better internal enablement/onboarding, and more reliable customer support experiences. All grounded in your organisations actual materials, anything from web pages, to call transcripts to earnings call recordings.. Worth a read if you’re thinking about AI as a durable capability—not a one-off chatbot project. #AgenticRAG #RAG #MarTech #MarketingOps #GenAI #EnterpriseAI #ContentStrategy #AIQuality

6

Our team at Turing just wrapped a project with Salesforce AI Research, putting models through Olympiad‑level maths. 🧠 We built a benchmark, Hard2Verify, that checks every step of the reasoning, not just the final answer. Turns out even top models make mistakes along the way. It’s exactly the kind of problem we enjoy solving - getting AI to actually think, not just guess!

9

Just getting started with LLMs? Don’t let the jargon slow you down. This 5-part LLM Glossary is your shortcut to mastering the fundamentals👇 1️⃣ Model Types • Foundation Model: Pretrained on large-scale, diverse datasets to learn general capabilities. • Instruction-Tuned Model: Further trained to follow user instructions precisely. • Multi-modal Model: Handles multiple input/output formats like text, images, and audio. • Reasoning Model: Optimized for logic, problem-solving, and step-by-step thinking. • Small Language Model (SLM): Lightweight models for fast, efficient, on-device tasks. 2️⃣ Training & Fine-Tuning • Pretraining: Initial learning phase using massive, diverse data sources. • RLHF (Reinforcement Learning with Human Feedback): Aligns model behavior with human preferences. • DPO (Direct Preference Optimization): Uses ranked preferences instead of rewards for training. • Synthetic Data: AI-generated data used to supplement real datasets. • Fine-Tuning: Targeted retraining on specific domains or tasks. • LoRA / QLoRA: Techniques for efficient low-resource fine-tuning. • Guardrails: Rules or filters applied to enforce safety, ethics, and compliance. 3️⃣ Prompt Engineering • System/User Prompts: Define model behavior and user context in a session. • Chain-of-Thought (CoT): Prompts that guide the model to reason step-by-step. • Few-Shot / Zero-Shot Learning: Demonstrating tasks with few or no examples. • Prompt Tuning: Optimizing prompts via training for specific outcomes. • Context Window: The total number of tokens the model can "remember" per interaction. 4️⃣ Inference & Generation • Temperature: Adjusts output randomness; lower values are more deterministic. • Max Tokens: Limits the number of tokens in a generated response. • Seed: Controls reproducibility of outputs. • Latency: Time taken by the model to respond. • Hallucination: When the model generates factually incorrect but plausible-sounding information. 5️⃣ Retrieval-Augmented Generation (RAG) • Retrieval: Fetching relevant data from external sources before generation. • Semantic Search: Search based on meaning, not just keywords. • Embeddings: Vector representations of text used for similarity matching. • Chunks: Dividing documents into smaller segments for better retrieval. • Vector Databases (VectorDBs): Stores embeddings for fast and accurate search. • Reranking: Rescoring retrieved documents to prioritize relevance. • Indexing: Structuring data for efficient retrieval during generation. 🌟 Follow the AIKaDoctor (Free AI & Data Science Resources) channel on WhatsApp: https://lnkd.in/dCTCEKKc 📌Follow Habib Shaikh For more such content.

232

36 (na) Komento

There’s a quiet irony in today’s “AI agent” debate. As The Information https://bit.ly/4avx4el reports, trust in generative AI is declining in enterprise settings. One reason: we’re asking agents to behave like deterministic systems. Most business workflows are algorithms: structured, repeatable, predictable. AI is excellent at building those workflows (writing code) and operating within them (summarizing, extracting, classifying). But the workflow itself is often better left deterministic. Agents aren’t computers, and shouldn’t be expected to act like them. The future is hybrid: workflows are deterministic; the work inside them often isn’t.

23

10 Komento

OpenAI’s Kevin Weil on the Future of Scientific Discovery Takeaways from a fireside chat between the OpenAI VP of Science and a16z speedrun visiting partner Sam Shank 1) The 5% → 80% jump. Models go from “this will never work”   to “it works 10% of the time”   to “why would I ever do this manually again?” That middle phase is where companies get built. Most teams wait for 80%.   The builders move at 10%. I’ve seen similar curves in autonomy and robotics. Early iterations look rough. Then something clicks. Then it becomes the roadmap. 2) Robotic labs + reinforcement loops. Model thinks.   Runs simulations.   Dispatches experiments to physical labs.   Gets real-world data back.   Refines. That’s reinforcement learning stepping into the physical world. When that loop tightens, discovery speed compounds. Materials. Drug development. Energy systems. Robotics. 3) Background agents. The mindset shift: always have something running. Before a meeting.   Before sleep.   During “dead time.” Treat idle time like idle compute. Last point that founders should internalize: Stop trying to solve complex flows with one giant prompt. Serious teams orchestrate. Planner model. Specialist models. Verification layer. Feedback loop. That’s architecture. Not prompt tricks. We’re in the 10% phase on a lot of capabilities right now. That’s the window.

2

We’ve moved beyond simple autocomplete—GenAI, agents, and chat-driven workflows are changing how real software gets built. This book - “Vibe Coding” by Gene Kim and Steve Yegge is about turning AI from cool demos into production-grade engineering. It shows how to work with AI effectively, put the right guardrails in place, and maintain quality, security, and engineering discipline while moving much faster. Perfect holiday read if you’re serious about how to use AI like a pro, not just why. 2026 is coming—time to sharpen the tools.

32

2 Komento

For #AI agents to work, focus on business outcomes, ROI not technology by Larry Dignan from Constellation Research, Inc. #AIF2025 - Check out the convo between Boomi @Chris Hallenbeck and @Ray Wang

3

Signs on Paper: How can we use foundation models to make sense of both modern and historical, and culturally rich sign language print dictionaries? We report on early eplorations in this poster to be presented Friday Sept 12 at CLIN25. Comments and ideas welcome! Nargess Asghari, Victoria Nyst and Peter van der Putten. Signs on Paper: exploring multimodal foundation models for classification of still images of signs in print dictionaries. Poster at Computational Linguistics in The Netherlands (CLIN35), Leuven, Sept 12 2025.

18

3 Komento

#AI coding #agents are changing how I evaluate vendors. A strong argument for products is if they can be configured purely through code and #API / #MCP. Not because they have their own AI features - but because they let your AI agent do the work. Lightdash is a great example. It is not the most mature BI tool, but probably one of the most AI-ready - because they made all data and configuration accessible via API and gave AI agents the context to act on it, so Claude Code can end-to-end create dashboards by modelling data in the data warehouse and then build the visualizations. If your product is not API-first, AI cannot bring in outside context or act autonomously within it. That is becoming a real competitive disadvantage. https://lnkd.in/e6Gp3MGj

53

14 (na) Komento

"Graphs make provenance first-class. Instead of burying "where did this come from?" inside a dense vector, a graph records sources, activities, and agents as nodes and edges you can query..." https://lnkd.in/efAdV9dx

26

Modernizing in regulated environments requires more than speed; it requires control. Today, we announced Nintex K2 (5.9), strengthening customization, accessibility, and integration capabilities for organizations running complex, on-premises workflows. As data sovereignty and compliance demands grow, automation platforms must adapt to enterprise requirements, not the other way around. K2 (5.9) continues our commitment to secure, flexible, enterprise-grade orchestration. Read the full announcement: https://lnkd.in/gyuExeNA #Nintex #K2 #ProcessAutomation #AI

24

The better the data, the smarter the Agent. Very exciting news: “In a world where every database speaks a different dialect of SQL, Waii’s metadata knowledge graph will create a unified semantic layer, enabling users and AI agents to understand and communicate with any data source while removing technical barriers to critical insights. We built Waii to close the gap between what people want to ask and what their data systems can understand,” said Gunther Hagleitner , co-founder and CEO, Waii

199

3 Komento

🧠 When Microsoft announced an MCP (Model Context Protocol) server for Docs, I wondered: How does it compare to Knowledge Sources when building #copilotstudio #agents that rely on web content? 🤖 In my latest article at Forward Forever blog, I explore both approaches and why comparing their results is essential when your goal is to generate answers from online information. 📖 Read more here: https://lnkd.in/dr3Hz6HZ

127

14 (na) Komento

Feeling grateful (and a bit humbled) to be surrounded by so many talented builders at actAVA: Kevin Riley Frank Wang Deon Metelski Steve Brown Joon Lee Leon Qi 🧠 As AI agents take on more autonomy in our day-to-day work, the gating factor in healthcare isn’t “capability”—it’s compliance to the workflow plan and instructions. That’s the only way we can safely let AI own repetitive tasks end-to-end. And as AI regulations expand—and clinicians and operators become more aware of how often models can be wrong—red teaming shifts from optional to foundational. You need an engine that continuously probes agents with adversarial scenarios, verifies they follow the right steps, and catches subtle failures before they ship. Autonomy in healthcare should be earned through evidence: evals, audits, and relentless testing. That’s what we deliver with Kora — a platform for healthcare AI agents with continuous evaluation, red teaming, and compliance assurance. Read more about our technology in the blog 👇. Any questions, let us know!

23

5 Komento