About the Role
Lead the effort to get Honestum’s intelligence on device. KPI's include: Architecture Review, Sprint Execution, UAT and Technical Handover
Requirements
What You Do
Read product requirement documents and create solution options for each of the L2s with supporting technical architecture. Provide an LOE (in terms of hours) for each epic based on the user stories within it. - Lead the conversion and optimization of large-scale transformer models (≥1B parameters) into Core ML formats (.mlmodel / .mlpackage) using corem ltools. Design and implement quantization workflows (INT8 and lower‑bit) via coremltools.optimize.coreml, balancing performance and accuracy. Profile and tune Core ML inference on Apple Silicon (ANE, GPU) for modern iPhones using Instruments (Core ML, Metal Trace), targeting minimal latency and optimal memory/throughput. Integrate Core ML models into production-grade Swift applications—handling model I/O, asynchronous execution, and robust error handling. - Write clean, production‑quality Swift code to load, manage, and invoke Core ML models with full error handling.
Attend grooming calls & sprint demos.
Collaborate closely with ML researchers and iOS engineers to align on requirements, benchmarks, and integration workflows- Ensure UAT standards are met:
Inference Latency: Meets or exceeds target ms/token on designated iPhone models
Quantization Accuracy: Maintains ≥ X% of floating‑point baseline post‑quantization
Throughput: Achieves specified tokens/sec benchmarks under real‑world load
Integration Stability: Zero critical runtime failures in staging and production builds
Deployment Velocity: Timely delivery of optimized models into app releases
Debug issues found during user acceptance testing (UAT).
Document feature set by L2 after Product Sign Off.
What You Have
Deep Core ML Optimization Expertise: Proven track record converting and fine‑tuning transformer models (>1B params) for Core ML
Quantization Mastery: Hands‑on experience with INT8 quantization (lower‑bit techniques a plus) via coremltools.optimize.coreml
Apple Silicon Performance Tuning: Demonstrable proficiency profiling and optimizing inference on ANE/GPU for low latency and high throughput
Swift Integration Proficiency: Strong ability to embed Core ML models in Swift/iOS apps, covering model I/O, async execution, and error management
Transformer Architecture Understanding: Solid grasp of transformer internals to debug conversion and performance challenges efficiently