About Mecka AI
Mecka AI is building the data infrastructure layer for robotics and embodied AI.
We partner with leading AI labs and robotics companies to deliver high-quality, real-world datasets used to train, evaluate, and deploy robotic systems. Our work sits directly between research, data, and real-world execution — where model performance is dictated by data quality.
The Role
We are looking for a VLA & LLM Engineer to serve as the primary engine for our large model training efforts. Your role is twofold: first, you will own the automated labeling system by training and fine-tuning LLMs and VLAs to process our data at scale. Second, you are the ultimate validator—you must prove that Mecka’s data actually works by training high-performing models that demonstrate superior results in real-world tasks.
This is not a theoretical research post. You will own the full process: from writing the math for a new architecture to debugging the production code that ships it. We don't hand you a finished system to work on; you are responsible for getting the models working from the start and meeting tight product deadlines.
What You’ll Work On
Automated Labeling & Pipeline Intelligence (Primary Focus)
Training the Pipeline: Train and fine-tune Large Language Models (LLMs) and Vision-Language-Action (VLA) models to automatically label, segment, and describe massive robotics datasets.
Knowledge Application: Use your ML expertise to improve our current processing pipelines. You will apply what you learn from training failures to refine how data is ingested, cleaned, and represented.
Semantic Reasoning: Develop systems where LLMs reason about high-level task goals while VLAs ground those goals in low-level sensory and action data.
Data Validation (Proving the Product)
The "Proof" Phase: You are responsible for proving that Mecka’s data is the best in the world. You will do this by training models that achieve state-of-the-art performance in both simulated and physical environments.
Full-Cycle Ownership: You are in charge of the full process. You will handle the setup, training, and debugging of the models. If a robot isn't performing, you are responsible for diagnosing if it’s an architectural flaw or a data gap.
Performance Benchmarking: Design experiments to show exactly how Mecka's data improves model generalization and failure-mode recovery.
Cross-Functional Support
Assisting All Teams: Apply your deep understanding of model mechanics to assist all engineering and research functions.
Production Debugging: Write, test, and debug production-grade code. You are motivated by shipping a functional product by the deadline, not just generating metrics for a paper.
Who You Are
Required Background
Education: A PhD or a Research-based Master’s degree in Computer Science, Robotics, Machine Learning, Physics, Mathematics, or a related quantitative field.
Proven Research History: A strong track record (publications at conferences like NeurIPS, ICRA, CVPR, or significant open-source contributions) demonstrating your ability to train and innovate within LLM or VLA frameworks.
Mathematical Rigor: You understand the first principles of transformers, multi-modal embeddings, and scaling laws. You don't just "tune" models; you understand the underlying math.
Production Engineer: You are an expert at writing and debugging production-grade code. You are motivated by shipping a functional product to a deadline rather than just publishing a paper.
Full-Stack ML Execution: You have experience taking a model from a blank script to a deployed, working system (in sim or on hardware).
Strong Signals:
Tenacity: You are comfortable being handed a non-functional system and taking total ownership until it works.
Architectural Intuition: You can look at a training curve and intuitively understand which architectural or data-level changes are needed.
Builder Mentality: You are research-driven but pragmatic. You want to see your research move real robots and solve real-world data problems.
Why This Role
Work directly with real-world robotics data at scale
Influence how leading robotics teams collect and use data
Apply your VLA/LLM expertise to real-world datasets that are larger and richer than almost any academic lab.
Compensation Range: $150K - $250K
