Leeroo

Meet SuperML: A plugin that gives you ML engineering superpowers. When connected, your agent handles the full ML pipeline. Give it a task, and it: Researches: Surveys the latest approaches to build a solid hypothesis. Plans: Drafts a sequential execution plan tailored to your hardware. Verifies: Checks your hyperparameters and configs before you burn compute. Debugs: Traces the actual root cause of errors and divergence. Iterates: Tracks hypotheses against results to define the next experiment. Agentic Memory: Carries your hypotheses, hardware specs, and lessons learned across sessions so your agent never repeats a failed experiment. Every single skill is grounded in real references. SuperML is powered by Leeroopedia, a massive knowledge base of ML docs, best practices, and framework guides. +60% Performance Boost: We tested SuperML on 38 highly complex ML tasks like Distributed Pretraining, DPO/GRPO Alignment, Multimodal RAG, Multi Agent Building, Synthetic Data Generation, and Prompt Optimization and saw a +60% performance boost over baseline coding agents. For enterprise teams, SuperML also ships with proven ML tasks for forecasting and planning, fraud and anomaly detection, customer analytics, recommendation systems, document intelligence, and customer service automation. 👇 Links to the GitHub repo and Enterprise access are in the first comment. #ML #AgenticAI #CodingAgent

🔥 Meet SuperML: A plugin that gives you ML engineering superpowers. When connected, your agent handles the full ML pipeline. Give it a task, and it: - Researches: Surveys the latest approaches to build a solid hypothesis. - Plans: Drafts a sequential execution plan tailored to your hardware. - Verifies: Checks your hyper-parameters and configs before you burn compute. - Debugs: Traces the actual root cause of errors and divergence. - Iterates: Tracks hypotheses against results to define the next experiment. - Agentic Memory: Carries your hypotheses, hardware specs, and lessons learned across sessions so your agent never repeats a failed experiment. Every single skill is grounded in real references. SuperML is powered by Leeroopedia, a massive knowledge base of ML docs, best practices, and framework guides. +60% Performance Boost: We tested SuperML on 38 highly complex ML tasks like Distributed Pretraining, DPO/GRPO Alignment, Multimodal RAG, Multi Agent Building, Synthetic Data Generation, and Prompt Optimization and saw a +60% performance boost over baseline coding agents. For enterprise teams, SuperML also ships with proven ML tasks for forecasting and planning, fraud and anomaly detection, customer analytics, recommendation systems, document intelligence, and customer service automation. 👇 Links to the GitHub repo and Enterprise access are in the first comment. #ML #AgenticAI #CdingAgent

Meet Grokipedia for AI/ML: The Brain that turns Generalist Agents into ML Experts. AI coding agents are great for general dev, but they hit a wall on complex AI/ML frameworks and data pipelines. Meet Leeroopedia, the Grokipedia for AI & ML best practices, built by AI for AI. It was built by the Leeroo Continuous Learning System, which spent thousands of hours ingesting and learning from 1,000+ top–tier ML/AI resources to form a transparent, Wikipedia-style knowledge base. It’s continuously updated by both AI and human engineers. When connected to coding agents, Leeroopedia MCP's gains speak for themselves: - 🚀 ML Inference Optimization: +17% relative speedup when writing complex CUDA and Triton kernels. - ✅ LLM Post-Training: +15% improvement in IFEval strict-prompt accuracy, with a +17% boost in serving throughput. - 🔍 Self-Evolving RAG: Built a RAG pipeline from scratch 16% faster, with a +13% improvement in F1@5 score. - 🥇 Agentic Workflows: Achieved an +18% improvement in customer support triage accuracy, processing queries 5x faster. Plug it into your IDE to build faster, more robust ML/AI systems: - Connect the MCP: Plug Leeroopedia into Cursor or Claude. - Give Your Prompt: Ask it to build or optimize your complex ML, Data, or AI task. - Watch it Build: Get highly optimized, state-of-the-art code the first time. 👇 Links to the Wiki and MCP Server are in the first comment! #MCP #AgenticAI

We’re sharing benchmark results on two of the hardest long-horizon, execution-grounded benchmarks: #1 on MLE-Bench (Kaggle-style ML / data science engineering), among open-source, reproducible systems #1 on ALE-Bench (AtCoder long-horizon algorithmic discovery) These results were produced with KAPSO: Knowledge-grounded Autonomous Program Synthesis and Optimization. The core idea is optimization-first codgen iteration under evaluation: ideate → edit/synthesize → execute → evaluate → learn, with reproducible experiment provenance and retrieval of reusable domain knowledge/principles/implementations/heuristics to avoid repeating failure modes. 💻 Code: https://lnkd.in/dr7gZW4w More updates coming soon with additional use cases. Plus, we’re launching Leeroopedia: A "best practices" wiki built by AI, for AI. 📚 Leeroopedia: https://leeroopedia.com/