Data Science Dojo engagement report

Best tweets

• Jul 24, 202652x their median

  💡 The AI industry has a naming problem. Five terms in five years. • 𝐏𝐫𝐨𝐦𝐩𝐭 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 (2022): Crafting the instructions you give a model in one exchange. • 𝐂𝐨𝐧𝐭𝐞𝐱𝐭 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 (2025): Managing everything the model knows beyond the prompt: retrieved documents, memory, tool definitions, session history. • 𝐇𝐚𝐫𝐧𝐞𝐬𝐬 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 (𝐅𝐞𝐛 2026): The constraints and gates that keep an agent in bounds. A good prompt won't stop an agent from rewriting your entire codebase if nothing architecturally prevents it. • 𝐋𝐨𝐨𝐩 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 (𝐉𝐮𝐧𝐞 2026): Designing the plan-execute-verify cycle an agent repeats, and deciding what triggers each step and what counts as done. • 𝐆𝐫𝐚𝐩𝐡 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 (𝐉𝐮𝐥𝐲 2026): Wiring multiple agent loops together through nodes, edges, and shared state, instead of relying on one agent to handle everything sequentially. Lay them out like this and the picture gets clearer. Prompt, context, and harness engineering are genuinely different layers of the same single-agent problem. Loop and graph engineering sit much closer together, a graph is just what happens when one loop isn't enough and you need several coordinating. And here's the part worth sitting with: harness to loop to graph is three renamed "eras" in about five months. That's a pace no other engineering discipline moves at, which says something about how much of this is new architecture versus new vocabulary. We found 6 frameworks that were already doing "graph engineering," nodes, edges, shared state and all, before the term existed. Full breakdown on the 𝐛𝐥𝐨𝐠, link in the 𝐜𝐨𝐦𝐦𝐞𝐧𝐭𝐬. #AIengineering #AgenticAI #LLMEngineering #AIagents #MultiAgentSystems #LangGraph

  339648818K viewsView on X
• Jul 27, 202614x their median

  📢 The retrieval step in a RAG system is only as good as the chunks feeding it — and most teams don't think carefully about chunking until something breaks. The way you split your documents determines what your retrieval system can find, what your LLM actually sees, and whether the answer it generates reflects the full picture or just a fragment of it. Most people treat chunking as a one-time setup decision — pick a size, move on. But every strategy makes a different tradeoff between how precisely a chunk can be retrieved and how much reasoning context it carries. Optimize for one without thinking about the other and you'll either get the right chunk with half the meaning, or a chunk rich enough to answer the question that never gets retrieved at all. There's no universal right answer here. There's only understanding what each approach actually does — and picking the one that fits what your system needs to do. This carousel walks through 5 chunking strategies, what each one gets right, and where each one quietly breaks down. #RAG #LLM #AIEngineering #VectorSearch #AgenticAI #RetrievalAugmentedGeneration

  9119307.9K viewsView on X
• Jul 23, 20267.4x their median

  🚨 Cursor just launched 𝐂𝐮𝐫𝐬𝐨𝐫 𝐑𝐨𝐮𝐭𝐞𝐫 in early access, an automatic model-routing layer. Cursor says it delivers frontier-quality coding results at 60% lower cost than routing every request to Opus 4.8, with no measured quality drop in internal testing. The detail that matters is what triggered this: • OpenAI rolled out hard spend limits across all API accounts the same week • Model routing is becoming the default way teams control cost on high-volume coding and agent workloads, not a manual optimization • Router picks a cheaper model automatically per request instead of leaving that choice to the developer, which only works if the routing logic reliably detects when a task actually needs the expensive model If you're running Cursor on a team plan, turn on Router in early access and compare your actual spend and output quality over a week before trusting the 60% number on your own codebase. We broke down how the Router classifier actually works, the real early-access savings range (30-50%, not just the 60% headline), and what to watch next. Full breakdown linked in comments. #CursorAI #ModelRouting #AICodingTools #DevProductivity #AIEngineering

  455907.7K viewsView on X
• Jul 29, 20266.6x their median

  🚨 Five days. 40 hours. One 𝐋𝐋𝐌 𝐚𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧 you'll build yourself. That's the shape of our next LLM Bootcamp, running August 24-28, 2026. We're not stopping at prompting basics: this goes into transformers and attention mechanisms, vector databases, fine-tuning, and the RAG challenges that trip up most production systems. By day five, you're evaluating your own 𝐚𝐠𝐞𝐧𝐭𝐬 and shipping a real build. 12,000+ alumni and 2,500+ companies have gone through this program. If you've been meaning to go from "I use LLMs" to "I can build with them," August is your window. Grab your seat before the cohort fills. Link in the comments! #LLMBootcamp #AgenticAI #RAGSystems #AIEngineering #FineTuning

  399505.6K viewsView on X
• Jul 23, 20266.3x their median

  🚨 Five days. 40 hours. One 𝐋𝐋𝐌 𝐚𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧 you'll build yourself. That's the shape of our next LLM Bootcamp, running August 24-28, 2026. We're not stopping at prompting basics: this goes into transformers and attention mechanisms, vector databases, fine-tuning, and the RAG challenges that trip up most production systems. By day five, you're evaluating your own 𝐚𝐠𝐞𝐧𝐭𝐬 and shipping a real build. 12,000+ alumni and 2,500+ companies have gone through this program. If you've been meaning to go from "I use LLMs" to "I can build with them," August is your window. Grab your seat before the cohort fills. Link in the comments! #LLMBootcamp #AgenticAI #RAGSystems #AIEngineering #FineTuning

  417207.5K viewsView on X
• Jul 29, 20265.0x their median

  LLM Wikis take a different approach to agent memory. Instead of just retrieving raw chunks like RAG (and then forgetting them), the agent actually curates what it learns: it reads new information, reconciles it against what it already knows, and rewrites its own knowledge into clean, current pages — much like a human maintaining a living wiki. In this session, Izma Aziz, Senior Software Engineer - Generative AI and LLMs at Data Science Dojo, breaks down how LLM Wikis actually work — and why they matter for agents that need to remember. Here's what we'll cover: 🔹 What LLM Wikis are, and how an agent maintains its own knowledge base 🔹 How they differ from RAG, file search, and chat memory 🔹 Where they pay off — fewer tokens, lower cost, cleaner knowledge 🔹 How knowledge gets created and updated, in a live build 🔹 What makes agent memory reliable in production If you're building or maintaining LLM-powered agents and want a real alternative to RAG for long-term memory, this one's for you. 📅 August 5, 2026 | 1:00 PM PT Register here: https://t.co/YXfsSm8qKb #LLMWikis #AIAgents #AgentMemory #GenerativeAI #LLM #LangGraph #DeepAgents #RAG #AIEngineering #MachineLearning #ArtificialIntelligence

  335205.3K viewsView on X
• Jun 30, 20264.4x their median

  The July 14th cohort is coming up soon. 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 July 14th looks like. 3 hours a week, structured to take you from foundations to deployment. Register now before it's too late: https://t.co/gOKkz5sjhv #agenticai #rag #aibootcamp #vectordatabases

  275305.2K viewsView on X
• Sep 4, 20263.0x their median

  Prompt injection isn't the only way an agent goes rogue. Sometimes it's just doing exactly what it was told, with permissions that were too broad to begin with. Join Dan Ndombe, Staff Developer Success Advocate at Docker, for a hands-on session on running LLM agents safely using Docker Sandboxes. We'll spin up a sandboxed agent runtime, give it a real task, then tighten the policy around it so it can only do what it's meant to do. In this session, you'll learn: - How Docker Sandboxes isolate an agent's filesystem, network, and process access from the host - How to hand a sandboxed agent a real task, end to end - How to read and tighten a sandbox policy: what to allow, what to block, and why - Common ways "sandboxed" agents leak permissions anyway - A reusable containment pattern for your own agent prototypes No prior Docker experience required. We'll build live and take questions. 📅 Wednesday, September 16, 2026 | 11:00 AM – 12:00 PM PT 🎙️ Dan Ndombe, Staff Developer Success Advocate, Docker 📍 Virtual 👉 Register here: https://t.co/zceI23qOWX #AgenticAI #AIAgentSecurity #Docker #datasciencedojo

  177003.8K viewsView on X
• Jul 22, 20263.0x their median

  💡 Getting an LLM to produce a coherent response in a demo takes an afternoon.   Getting it to behave consistently across edge cases, stay within cost, and not hallucinate on the inputs that matter — that's where teams spend months they didn't plan for.   This carousel covers the 10 𝐬𝐤𝐢𝐥𝐥𝐬 that actually close that gap: from RAG pipelines and vector databases to fine-tuning with LoRA, evaluation frameworks like RAGAs and G-Eval, and agentic systems built on MCP.   Each one maps to a module in our 𝐋𝐚𝐫𝐠𝐞 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐌𝐨𝐝𝐞𝐥𝐬 𝐁𝐨𝐨𝐭𝐜𝐚𝐦𝐩 — 40 hours, in-person and online, starting August 24th. If you want to build LLM applications that hold up in production, this is the structured path to get there.   Link to the curriculum in the comments. #LLMBootcamp #LargeLanguageModels #AIEngineering #RAG #AgenticAI #LLMDevelopment

  174304.2K viewsView on X
• Aug 19, 20262.9x their median

  LLM Wikis: How AI Agents Build a Second Brain That Never Forgets https://t.co/0VF3GymTWT

  162416.1K viewsView on X

Ranked by total interactions across everything we have tracked for this account, which is a longer history than the 30-day window the rates above use. The multiple compares each post to this account's own median.

Keep going