Data Science Dojo

The May 5th cohort is 6 days away. 10 weeks. 12,000+ alumni. Spots are filling fast. The Agentic AI Bootcamp is back and it covers the full stack of what matters right now. Week 1, you're wrapping your head around how Transformers actually work. Week 3, you're building with LangChain and setting up your first vector database for a RAG pipeline. By week 5, you're not just prompting. You're engineering context and designing systems using Agentic design patterns used in production. Week 7 hits and you're implementing Model Context Protocol and learning how agents actually talk to each other. Week 10? You ship a multi-agent LLM application. From scratch. In 10 weeks. That's what the Agentic AI Bootcamp cohort starting May 5th looks like. 3 hours a week, structured to take you from foundations to deployment. Register now before it's too late: hubs.la/Q04dShn50 #agenticai #rag #aibootcamp #vectordatabases

🚀 Your AI agent can think… but can it act? Join “Skills, Tools & MCP – Connecting Your Agent to the World” with Kwasi Ankomah (Lead AI Architect, SambaNova Systems) 🗓 13 May, 2026 | ⏰ 11:00 AM Learn how to: • Build custom tools • Connect agents using MCP • Create reusable skills • Turn ideas into real-world, production-ready AI systems 👨‍💻 Perfect for AI engineers & developers working with LLMs 👉 Register now:hubs.la/Q04flK7q0 #AIAgents #AgenticAI #AIEngineering

is harness engineering the next real lever for coding agent performance, or does it eventually hit a ceiling that only better models can break through?

📢 Most coding agent research focuses on making the model smarter. This paper argues the bottleneck is elsewhere — in the harness. A new paper from Fudan, Peking University, and Shanghai Qiji Zhifeng introduces Agentic Harness Engineering (AHE): a closed-loop system that automatically evolves all the scaffolding around a coding agent — tools, middleware, memory, system prompt — without touching the base model. The core insight: harness evolution fails not because the optimizer is weak, but because it can't see clearly enough to act. AHE fixes this through three observability layers: 1️⃣ Component observability: Every editable part of the harness (tools, middleware, skills, memory) exists as an explicit file, so the optimizer has a clean action space and every edit is attributable to a single component. 2️⃣ Experience observability: Instead of feeding the optimizer millions of raw trajectory tokens, AHE distills rollouts into a layered evidence corpus with per-task failure analysis. The optimizer reads structured root causes, not logs. 3️⃣ Decision observability: Every harness edit ships with a self-declared prediction of which tasks it should fix and which it might break. The next evaluation round either confirms or rolls back the change. After 10 iterations from a minimal bash-only seed, AHE lifts pass@1 on Terminal-Bench 2 from 69.7% to 77.0% — surpassing both human-designed harnesses and self-evolving baselines like ACE and TF-GRPO. The evolved harness also transfers: on SWE-bench-verified it achieves the highest aggregate success rate at 12% fewer tokens than the seed, and across three alternate model families (DeepSeek, Qwen, Gemini) it yields +5.1 to +10.1 pp gains without re-evolution. One finding worth noting for anyone building agents: the system prompt alone regressed performance when swapped in isolation (-2.3 pp). Tools, middleware, and long-term memory each independently improved it. The harness components most people don't think to optimize are exactly where the gain lives. #agenticai #llmengineering #aiengineering #harnessengineering

⏰ One day left to register. 📢 Most people learning agents hit the same wall: strong LLM knowledge, but no mental model for how a real system holds together. You can pick a framework. You can write a prompt. What's harder is designing something that reasons reliably, manages state, uses tools without breaking, and behaves predictably outside a demo environment. That's the specific gap this bootcamp is built around — not concepts in isolation, but how reasoning, memory, tool use, and system design connect in practice. Starting from first principles, working up to production-style agentic systems. If things almost make sense and you want them to fully click, register now for the bootcamp starting May 5th, 2026: hubs.la/Q04fhnW50 #AgenticAI #LLMEngineering #AIEngineering #DataScience

📢 Prompt engineering is not the skill. Context engineering is. Most people treat a prompt like a search query — one box, one question, send. But an LLM isn't reading just your message. It's reading a stack: → The system prompt defining who it's supposed to be → Your actual instructions → The entire conversation history leading up to this moment If any layer is weak or conflicting, the output suffers and you blame your wording instead of your architecture. Context engineering fixes that. It's the practice of deliberately shaping all three layers: setting a precise system role, pruning irrelevant history, injecting the right external context before you ask for anything. The people building reliable AI workflows aren't writing better prompts. They're designing better context. Swipe to see all 3 layers broken down. #llm #aiengineering #contextengineering #promptengineering #generativeai

MCP and A2A are both agent protocols but they operate at completely different layers. MCP (Model Context Protocol) is about giving one LLM access to external tools. The model stays in the driver's seat throughout: it receives your query, decides which tools to call, gets the results back through an MCP Client, and assembles the final response. The MCP Servers in between are just standardized wrappers around raw APIs — Flight Booking, Google Calendar, whatever you need. One brain. Centralized control. A2A (Agent-to-Agent Protocol) is about coordination between autonomous agents. There's no single LLM managing everything — instead, an Orchestrator Agent delegates subtasks to specialized agents, each of which has its own tools, its own memory, and its own reasoning loop. The Orchestrator never touches the Flight API directly. It talks to a Flight Agent, which handles that domain entirely on its own, then reports back. The architectural difference matters when you're building: MCP works well when one model can reasonably handle the full task with tool access. A2A becomes necessary when subtasks are complex enough to need dedicated reasoning — or when you want agents that can operate in parallel without bottlenecking through a single model. Same scenario — "book me a flight and add it to my calendar" — but one architecture keeps control centralized, the other distributes it across specialists. If you're building agentic systems in 2026, understanding where this boundary sits will shape every design decision you make. #agenticai #aiarchitecture #mcp #google #anthropic #aiengineering #multiagentsystems

Week 5 context engineering is where this curriculum shines. This is what separates my working IoT systems from hallucinating demos. Most teams ignore these principles. ⚡️

Top AI Agentic Design Patterns Source: @DataScienceDojo #GenerativeAI #GenAI #LLM #LargeLanguageModel #AI #python #ml #BigData #datascientist #artificialintelligence #NaturalLanguageProcessing #data #AgenticAI #AppliedAI #Automation #machinelearning #deeplearning #AIAgents

@AnandButani It definitely sounds too good to be true!

@DataScienceDojo shipped with Kimi once when Claude's API costs spiked mid-sprint. the "agent swarm" mode sounded clean until three parallel tasks contradicted each other with no arbitration layer.

Something interesting happened the week we ran these tests. Anthropic spooked a lot of developers with a pricing page change — Claude Code appeared to be removed from the Pro plan. It was reversed within 24 hours, but the reaction was telling. People weren't just annoyed, they were actively scoping alternatives. Kimi kept coming up. Kimi K2.6 had just launched with a strong pitch: open source, 3-4x cheaper on the API, and three distinct execution modes including Agent Swarm — something no other model in this tier currently offers. We tested it against Claude Sonnet 4.6 across 4 developer tasks. Both models are good. The differences aren't dramatic, but they're consistent and they point toward two different philosophies about what "helpful" means for a developer. One leans toward completeness. The other leans toward production awareness. Depending on what you're building, that distinction matters more than any benchmark number. Full breakdown in the comments 👇 #kimik2 #claudecode #anthropic #kimiai #agenticai

@khalide_f1 Definitely! it could make or break your product.

@DataScienceDojo That split is the interesting part. Clever outputs vs production aware code. When you're building real systems, that difference matters way more than benchmark charts 👀

Effortless Integration: Connecting LinkedIn API to Azure Synapse / Data Factory for Centralized Marketing Analytics in Power BI buff.ly/48PvVG9 v/ @DataScienceDojo #DataScience #AI Cc @IanLJones98 @Fabriziobustama @chidambara09 @nincoroby @mvollmer1 @mikeflache @ahier

Le ROI de l'#IA agentique en 3 étapes : 1️⃣ Réduire les coûts. 2️⃣ Générer du revenu. 3️⃣ Créer ce qui était impossible avant. via @DataScienceDojo

⚡Most people blame the model when their AI agent underperforms. In reality, the problem is almost always the context around it. That’s where context engineering comes in. It’s built on 6 key components: • Instructions / System Prompt → Defines the agent’s role and rules • Long-Term Memory → Retains state across sessions • Available Tools → Determines what actions the agent can take • Structured Output → Keeps responses consistent and usable • User Prompt → The input being responded to • Retrieved Information (RAG) → Pulls in the right data at the right time When these are designed intentionally, your AI: → Stops hallucinating missing context → Makes better decisions → Actually works in real workflows — not just demos In this hands-on session with Kwasi Ankomah, Lead AI Architect at SambaNova, you’ll learn how to apply all 6 - and build a real, production-ready AI agent. Catch the full session here: hubs.la/Q04f3K1R0 #AgenticAI #LangGraph #AIAgents #AIEngineering #SambaNova

🚀 The Agentic AI Bootcamp is just days away and Luis Serrano will be teaching Module 2 on Transformer Architecture & Attention Mechanisms! Join Luis Serrano, Founder & CEO of Serrano Academy, as he breaks down how transformers work under the hood and why attention mechanisms are at the core of every modern AI system. 📅 Bootcamp starts May 5th, 2026 🎟️ Grab your spot now: hubs.la/Q04f2g3g0 #agenticai #aibootcamp #transformers #llm #aiengineering

Read the complete comparison with the feature table, all 4 task results, and our observations: hubs.la/Q04f0m0d0

Rethinking Knowledge Work in the Age of AI x.com/i/broadcasts/1…

🚀 Happening today in Seattle! Join us this evening for an in-person discussion on how AI is reshaping knowledge work, business models, and the future of professional services. We’ll explore what this shift means for leaders, teams, and professionals navigating the AI era. We’ll dive into: • how AI is reshaping business models • where human expertise still creates durable advantage • what real AI upskilling looks like today 📍 Location: 5010 148th Ave NE, Suite 200A, Redmond, WA 98052, United States After a short presentation, we’ll open the discussion for audience Q&A. 👉 Register here: hubs.la/Q04dXgBS0 Can’t make it in person? You can still join online - it will be livestreamed. #ArtificialIntelligence #FutureOfWork #AILeadership #KnowledgeWork

Large Language Models Bootcamp Information Session x.com/i/broadcasts/1…