Company Description
Blend is a premier AI services provider, committed to co-creating meaningful impact for its clients through the power of data science, AI, technology, and people. With a mission to fuel bold visions, Blend tackles significant challenges by seamlessly aligning human expertise with artificial intelligence. The company is dedicated to unlocking value and fostering innovation for its clients by harnessing world-class people and data-driven strategy. We believe that the power of people and AI can have a meaningful impact on your world, creating more fulfilling work and projects for our people and clients. For more information, visit www.blend360.com
Job Description
We are looking for a Data Scientist to drive performance optimization for our clients AI Engine. This high-impact role will tackle complex computational bottlenecks and contribute to the re-engineering of simulation, training, and post-processing pipelines for large-scale industrial AI models. The candidate will collaborate with a multidisciplinary team to deliver quantifiable improvements in speed, scalability, and memory efficiency.
Responsibilities:
Analyze and optimize AI engine code, focusing on removing performance bottlenecks in simulation, training, and post-processing workflows.
Refactor sequential code and nested loops into efficient, vectorized operations, leveraging advanced knowledge in linear algebra and matrix decomposition.
Diagnose and resolve computational inefficiencies related to GPU/TPU, non-vectorized data handling, and mixed framework operations (JAX, NumPy, Pandas).
Develop and implement solutions for ODE (ordinary differential equation) solvers, optimization algorithms, and batch processing strategies.
Lead root cause analysis for performance limitations and propose alternative algorithmic strategies (including MILP/LP, decomposition techniques, and alternative ODE solvers).
Guide remediation and refactoring efforts, including memory optimization, JIT compilation, and data type standardization.
Document improvements, monitor ongoing performance, and contribute to a roadmap for further scalability enhancements.
Qualifications
Masters in Mathematics, Operations Research, Computer Science, or related quantitative field.
Deep expertise in matrix decomposition, numerical optimization, and ODEs, with a demonstrated ability to apply these in real-world computation.
Strong proficiency in Python, with hands-on experience in JAX, and experience with GPU/TPU acceleration.
Prior exposure to optimization in scientific computing, especially in re-engineering or scaling data pipelines.
Familiarity with ensemble methods, batch and vectorized computation, and memory management in large datasets.
Ability to communicate technical findings and lead cross-functional discussions for codebase improvement.
Preferred:
PhD preferred
Background in large-scale industrial AI or simulation platforms.
Knowledge of MILP, LP decomposition, and alternative ODE solvers (e.g., diffrax, odeint).
Experience transitioning AI workloads from CPU to GPU/TPU environments.
Experience with JAX, PyTorch or TensorFlow as alternative computation frameworks.
Experience with CI/CD practices for ML/AI pipelines, profiling tools (e.g., cProfile, memory_profiler, jax.profiler), and performance benchmarking.