Core ML Engineer: Deep Learning Architecture
Role summary
Mecka is seeking an ML and Optimization Specialist to lead improvements in model architecture across all pipelines, focusing initially on converting frame-by-frame models to temporal inference for critical performance gains. This deeply technical, high-leverage role requires expertise in ML model architecture design, optimization, and debugging at the architectural level, with a strong background in temporal/sequential models and PyTorch. The specialist will also profile and optimize model performance, evaluate new techniques, and collaborate with cross-functional teams for deployment. The role offers significant ownership and impact in a well-funded robotics AI company.
## The Role
We're hiring an ML and Optimization Specialist to lead model architecture improvements across all of Mecka's pipelines.
Many of our current ML systems rely heavily on frame-by-frame models, but all of our data is inherently temporal. Your immediate focus will be converting and optimizing these models for temporal inference — a critical unlock for pipeline performance.
Beyond that, you'll be the go-to person for model-level debugging, architecture design, and optimization across the organization. This is a high-leverage, deeply technical role for someone who thinks at the architecture level.
## Responsibilities
### Immediate Priorities
- Temporal model conversion — migrate frame-by-frame models to temporal architectures that leverage sequential data
- Benchmark and validate temporal models against existing frame-based baselines
### Ongoing
- Lead model architecture improvements across all pipelines (CV, pose estimation, etc.)
- Tune and debug ML models at the model architecture level — not just hyperparameters, but structural decisions
- Profile and optimize model performance (latency, throughput, memory)
- Evaluate and introduce new architectures, training strategies, and optimization techniques
- Collaborate with CV, ML, and infrastructure teams to deploy improved models
## Who You Are
### Required Skills
- Deep expertise in ML model architecture design and optimization
- Ability to tune and debug models at the architecture level — diagnosing issues in attention mechanisms, loss landscapes, gradient flow, etc.
- Strong experience with temporal/sequential models (transformers, RNNs, temporal convolutions, state-space models)
- Proficiency in PyTorch (or equivalent) at a research-engineering level
- Experience optimizing models for production deployment
### Strong Signals
- Published papers or production experience with video understanding or temporal perception
- Experience with model distillation, quantization, or efficient inference
- Background in computer vision model architectures
- Experience converting or adapting pre-trained models to new domains/modalities
- Familiarity with ONNX, TensorRT, or similar inference optimization tools
### You Are
- Obsessed with model internals — you think in terms of architecture, not just training runs
- Highly systematic in debugging and benchmarking
- Able to move between research papers and production code
- A strong communicator who can explain architecture tradeoffs to cross-functional teams
## Why This Role
- Own the model architecture strategy across all of Mecka's pipelines
- Solve a critical temporal modeling challenge with immediate impact
- Work at the intersection of perception, robotics, and ML systems
- High ownership in a fast-moving, well-funded robotics AI company
Compensation Range: $160K - $250K