Leeroo recently launched!

‍

^{"Trainable AI agents that keep learning—and getting better at their job—every day."}

‍

Hi everyone 👋

Leeroo delivers trainable AI agents that learn much like human colleagues— from knowledge bases, human feedback, and even their own past experiences. That continuous learning bridges the gap between generic AI and the expert performance companies need. They’re starting with agents for data- and AI-teams.

‍

https://youtu.be/ONHTKTB7Dik

‍

Founded by Majid Yazdani, Arshad Shaikh & Alireza Mohammadshahi‍

‍

Team

The founders have worked together for 3 + years on various AI systems. Altogether the team has co-authoring 50 + papers in top AI venues, open-sourcing models downloaded 1.5 million+ times, and earning 3K+ GitHub stars.

Majid – Co-founder & CEO

PhD, EPFL (ML). Built and led multiple AI teams, created core technologies and AI-enabled products, published research articles in top-tier venues —some featured in media. He was a staff Scientist at Meta AI, where he co-developed a breakthrough technology called “Neural Databases” and drove LLM applied research that powered customer-support automations and feed relevance across Meta’s apps. Before that, he was staff Scientist at LinkedIn AI, led content-understanding, job-profile matching, and economic-graph representation learning. Former VP of Research & Technology at BYJU’S; earlier research posts at the University of Geneva and Idiap Research Institute.

Arshad – Co-founder & CTO

With an M.S. in Computer Science from Mumbai University, Arshad bridged research and production at BYJU’S AI Labs—building BADRI, a multimodal transformer–based recommender trained on 30 billion interactions. He also engineered an inference stack that serves over 10 million students efficiently. Prior to that, he was a deep-learning researcher at IIT Bombay.

Alireza – Co-founder & Chief Scientist

With a PhD in CS/AI from EPFL and a proven record across leading labs, he has repeatedly advanced LLM research and deployment: at Meta AI, he built an LLM-driven evaluation pipeline that boosted LLM performance on out-of-domain datasets; at Naverlabs, he compressed a state-of-the-art multilingual machine-translation model by 40x; at the University of Zurich, he introduced ensemble and contrastive decoding techniques that significantly lifted translation accuracy; and at the Idiap Research Institute, he devised methods for encoding structured data in LLMs, enhancing a range of NLP tasks. His open-source releases have collectively surpassed one million downloads on Hugging Face, and the underlying research has produced fifteen papers in top-tier AI conferences.

‍

❌ Problem

AI agents look impressive, but for specialized work they remain semi-experts—and you can’t effectively train them into experts. General-purpose agents can’t learn from their own experience, human guidance, or domain-specific materials, so they quickly plateau at basic capabilities. Data and AI tasks demand deep specialization that off-the-shelf agents simply don’t reach.

‍

✅ The Solution

Scaling pre-training to squeeze out marginally better foundation models has flat-lined. The real moat isn’t building demo agents, it’s training that agent to expert level on your private context. While agent scaffolding is commoditising fast, Leeroo’s on-prem, continuous-learning engine transforms your data, workflows, and tacit know-how into a compounding strategic asset—without leaving your VPC.

What makes Leeroo different:

• Continuous Learning Loop: Every interaction becomes training data. Agents digest codebases, docs, videos, self-discovery experiences, and your real-time feedback, then consolidate the lot overnight—so yesterday’s lessons fuel tomorrow’s results.
• Domain-Specific Reasoning Library: Symbolic search, specialized code synthesis, and deep data/ML expertise tackle real engineering problems directly, not through brute-force guessing.
• Multi-Agent Workforce: Multiple agents collaborating asynchronously on heavy jobs, could run entirely inside your VPC.

Leeroo's early customers already see their trained agents as high-value assets—assets that appreciate as the agents keep mastering more of the organization’s knowledge.

‍

The Ask

The founders are lining up pilot partners in banking, finance, and healthcare who need AI agents that respect data sovereignty yet deepen domain expertise over time.

Help them bring continuously learning AI to the industries that need it most.

‍

Learn More

^‍

^{🌐 Visit leeroo.com to learn more.}

^‍

^{🤝 If you know innovation leaders wrestling specifically with data and AI problems, they founders would love an intro.}

‍

^{👉 Contact them via email or book a demo.}  

^‍

^{👣 Follow Leeroo on LinkedIn & X.}

‍