SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Beltic is building real-time agent verification systems that run on the request path—meaning models must operate under strict production latency budgets with no room for offline batch processing or retries. This is a high-stakes ML engineering role where you'll own the full lifecycle of production models: from feature engineering and training through evaluation to low-latency, real-time serving.
You can work remotely from anywhere in the Americas (UTC-8 to UTC-3 time zones) or from the San Francisco office, with flexibility to choose your preferred arrangement.
The role focuses on building robust ML systems in adversarial domains. You'll tackle challenges like fraud detection, account integrity, and security—where threat models evolve and labels are scarce. The work spans anomaly detection, unsupervised learning, weak supervision, and building evaluation sets from scratch. You'll also engage with emerging areas like graph neural networks, behavioral modeling, LLMs, AI agents (tool use, function calling, MCP), and explainable ML with full model versioning and auditability.
This is a senior individual contributor role requiring deep production ML experience, not research or prototyping. You'll ship versioned, testable, reproducible code and collaborate closely with founders on high-impact problems.
Requirements:
- Years of production ML experience: building, evaluating, and deploying models (not research or notebooks)
- Strong Python and modern ML framework (PyTorch or TensorFlow)
- Solid SQL
- End-to-end production ML systems: feature pipelines, training, evaluation, low-latency real-time serving
- Comfort with low-label and cold-start problems, anomaly detection, unsupervised methods, weak supervision
- Data analysis, statistics, and experiment design
- Strong software engineering: versioned, testable, reproducible code
Strongly Preferred:
- Applied ML in fraud, risk, abuse, account integrity, or security
- Graph ML / GNNs, sequence or behavioral modeling, anomaly detection
- LLM / NLP work and familiarity with AI agents (tool use, function calling, MCP)
- Adversarial settings where threat models evolve
- Explainable, auditable, reproducible ML with real model versioning
- Knowledge of payment protocols (x402), agent identity (DID / VC), or crypto commitments (Merkle anchoring)
