Applied Machine Learning Engineer

Role Overview

Sentient Labs is seeking an Applied Machine Learning Engineer skilled at bridging machine learning research and robust production software development. This comprehensive engineering position requires the capability to interpret scientific literature, distill workable experiments, conduct rigorous evaluations, and convert findings into reliable, user-accessible systems.

Key Responsibilities

• Recreate and analyze research techniques utilizing openly accessible model weights and APIs.
• Create datasets, assessment tools, scoring approaches, calibration procedures, baselines, and experimental frameworks tailored for evaluation.
• Engage directly with model components such as weights, logits, hidden states, activations, APIs, and inference architectures as needed.
• Develop and enhance our evaluation infrastructure including executors, adjudicators, data persistence, orchestration of experiments, and reporting tools.
• Transform research procedures into user-facing product features with clear experiment configuration, execution, tracking, comparative analysis, and review processes.
• Examine behavior of verification mechanisms when models undergo modifications like fine-tuning, merging, quantization, distillation, safety alterations, or evasion attempts.
• Design controlled studies that effectively isolate meaningful signals from artifacts or confounding variables.
• Produce articulate technical reports differentiating empirical results, analysis, and theoretical hypotheses.
• Deliver production-grade systems integrating APIs, background processes, monitoring, quality assurance, and comprehensive documentation.

Candidate Profile

• Strong proficiency in Python programming coupled with practical experience using PyTorch and Hugging Face Transformers.
• Deep understanding of machine learning evaluation methodologies such as dataset construction, baseline development, metric analysis, calibration, error types, statistical confidence, and reproducibility.
• Capability to interpret scientific ML publications and implement foundational methods without exclusive reliance on preexisting libraries.
• Proven track record in building production software beyond prototyping environments, including API design, asynchronous task handling, database management, observability, testing, and deployment practices.
• Competence with open-weight model handling and insights into the mechanics of current large language model (LLM) inference systems.
• Ability to cross the backend-frontend divide, working with React and TypeScript to make intricate experimental data accessible and understandable to end-users.
• Excellent technical judgment in discerning the validity of experimental evidence and the underlying assumptions, avoiding superficial interpretations.
• High initiative and ownership mindset, proactively addressing challenges and driving projects without needing detailed supervision.
• Adaptability to the dynamic environment of a fast-paced startup, willing to engage across multiple functions as priorities shift.

Preferred Additional Experience

• Knowledge of techniques such as model provenance tracking, fingerprinting, watermarking, distillation detection, red-teaming, security evaluation, or interpretability.
• Experience analyzing model internals including activations, representations, probing, model hooks, logits, and hidden states.
• Familiarity with evaluation and inference platforms like DSPy, LiteLLM, Temporal, Ray, vLLM, and PostgreSQL/pgvector.
• Expertise in front-end technologies such as Next.js, React, TypeScript, data visualization, and dashboard creation for experiments.
• Skills in deploying and operating open-weight models on GPUs with considerations for latency, throughput, memory footprint, precision, and cost efficiency.
• Experience designing adversarial testing methodologies or evaluation frameworks resilient to intentional evasion efforts.

Success Milestones Within Six Months

• Successfully replicate at least one documented model provenance or verification methodology, thoroughly outlining its assumptions, capabilities, and limitations.
• Create a reproducible runner for model verification with version-controlled inputs, artifacts, metrics, and reporting mechanisms.
• Integrate a new verification workflow in Construct and enable its accessibility through the Eldros user interface.
• Conduct systematic experiments across various model types including base, fine-tuned, merged, quantized, and distilled models.
• Enhance the understanding of verification method effectiveness, failure modes, and root causes.
• Develop well-engineered, maintainable codebases, complete with tests, tooling, and documentation to support fellow engineers.

Exclusions

• This is not a pure research role limited to publishing papers or exploratory notebooks.
• Not a generalist role focused solely on model training or fine-tuning tasks.
• Does not limit to exclusively front-end or back-end engineering responsibilities.
• Rejects reliance on benchmark numbers without a deep understanding of their derivation.
• Requires flexibility to work across multiple technical layers rather than being confined to a single stack layer.

We seek a versatile engineer passionate about integrating research insights with product-grade, trustworthy experimental systems.