About SelfInspection SelfInspection is building a Vehicle Condition Intelligence Platform: the system of record for vehicle condition across a vehicle's entire lifecycle. Our AI turns fragmented inspection and handoff data into a trusted, continuously updated source of truth. Rental operators, fleets, dealerships, finance providers, insurers, transportation companies, automotive marketplaces and mobility platforms use it to manage risk, protect assets and reduce disputes. The role You'll own how our AI runs in production: training, evaluation, serving and monitoring for damage detection, tire analysis, image quality control and automated estimates. You'll combine our vision models with VLMs and LLMs into pipelines that are accurate, auditable and cheap enough to run on every inspection. This is a senior, hands-on IC role on a small team. You'll set ML engineering standards, mentor engineers, and work closely with computer vision, data, backend and product. Full-time, remote. What you'll build
• Production inference. Serve vision models and VLM pipelines on AWS with clear latency, throughput and cost-per-inspection targets.
• VLM and LLM pipelines. Prompted and fine-tuned stages for damage detection, verification and report generation, versioned in MLflow, with strict output schemas and no invented findings.
• Automated estimates. Models that turn detected damage into repair operations and cost estimates, measured against inspector estimates.
• Training and retraining. Reproducible training pipelines, experiment tracking, and scheduled retraining on newly labeled data.
• Evaluation. Offline harnesses against inspector ground truth, regression gates, shadow deployments and A/B tests before every release.
• Monitoring. Drift, data quality and accuracy monitoring in production, with alerts and a feedback loop into labeling.
• Cost and latency. Batching, quantization, model routing and caching that keep GPU and model-API spend per inspection under control.



What you bring
• 6+ years in ML engineering, with models you shipped and operated in production.
• Strong Python and PyTorch, with production-grade code, tests and CI.
• MLOps in practice: MLflow or similar for experiments, registry and deployment; Docker; AWS (SageMaker, ECS or EKS, Lambda, S3).
• Model serving at scale: Triton, TorchServe, vLLM or managed endpoints, with GPU sizing, batching and autoscaling.
• Production VLM/LLM work: prompting, structured outputs, evals and LoRA fine-tuning, and the judgment to pick a classical model when it wins.
• Evaluation rigor: metrics tied to business outcomes, calibrated thresholds and disciplined error analysis.
• ML monitoring: drift detection, data validation and alerting.
• Fluent professional English, spoken and written. Our team works in English every day.
• Clear communication of tradeoffs to product, sales and customers.
• BS/MS in Computer Science, Data Science, Electrical Engineering or a related field, or equivalent practical experience. Nice to have
• Computer vision: detection, segmentation and OCR models in production.
• Automotive repair estimating, insurance claims or pricing models.
• Agentic, tool-using LLM systems in production.
• Cost control across hosted model APIs (Gemini, OpenAI, Anthropic) versus self-hosted open models.
• Event-driven services: queues, webhooks and idempotent processing, integrated with TypeScript or NestJS backends.
• Working under ISO 27001 or SOC 2 controls. What success looks like in 6 months
• A measurable accuracy gain in damage detection against inspector ground truth, shipped to production.
• Every model and prompt release passes an automated regression gate.
• Latency and cost per inspection are tracked by stage and customer, with a reduction delivered.
• A first automated estimate model runs in shadow mode, measured against inspector estimates.
• Production monitoring catches drift and accuracy drops before customers do.

📌 Senior Machine Learning Engineer (México)
🏢 Self Inspection
📍 México