AI Compiler & Inference Engineer

Escrow-Tech is seeking an experienced AI Compiler & Inference Engineer to make customer AI inference workloads portable to the K3 CPU/NPU environment. The role focuses on model conversion, production-ready inference pipelines, numerical equivalence, reliable deployment, and safe handling of unsupported operators and models

Job Details

• Location: Overseas / Remote
• Job Type: Fixed-term contractor
• Duration: 15 October - 14 December 2026 (optional 2-4 week contract extension)

Key Responsibilities

• Profile the K3 NPU compiler/runtime environment, supported operators, model formats, quantisation modes, and deployment constraints
• Implement model intake and conversion for agreed launch model families, including supported ONNX/PyTorch-exportable models
• Design authorised quantisation and calibration workflows while preserving model provenance
• Develop operator-support matrices and deterministic fallback/rejection behaviour
• Create reference outputs and statistical/numerical equivalence tests
• Integrate model conversion into the One-Click planner and artefact manifest
• Optimise inference packaging and deployment without hidden vendor-tool dependencies
• Document customer limitations, performance considerations, and recovery/reconversion procedures
• Transfer model-support runbooks and test assets to the permanent engineering team

Requirements

• 5+ years of ML systems/inference engineering experience with production model deployment
• Strong experience with ONNX, PyTorch export, model compilers/runtimes, and quantization
• Experience with numerical validation and calibration workflows
• Familiarity with edge, accelerator NPU, or heterogeneous inference environments
• Strong Python/C++ and Linux deployment skills
• Experience developing CI-testable model pipelines
• Understanding of the difference between model conversion and retraining, including when original learned weights cannot simply be recreated

Project Milestones

• M1: Input/Runtime Assessment
• M2: First Model
• M3: Integrated Model Path
• M4: Support Matrix/Equivalence
• M5: Production Acceptance & Knowledge Transfer

Additional Details

• Global search, prioritising engineers experienced in model compiler/runtime deployment rather than general data science
• Optional 2-4 week contract extension

How to Apply

Apply below via the JobClave with your updated Resume/CV (include a cover letter)