About 10a Labs
10a Labs is an applied research and AI security company trusted by AI unicorns, Fortune 10 firms, and U.S. tech leaders. We focus on multilingual threat intelligence, abuse detection at scale, and cutting-edge AI red teaming.
The Role
We’re hiring an ML Engineer to design, train, and deploy production-ready ML systems, with a focus on applying traditional ML techniques to modern LLMs. This role offers end-to-end ownership of the ML lifecycle—covering data pipelines, training, evaluation, deployment, and continuous improvement.
Responsibilities
-
Build and deploy multi-stage classification systems with high recall, precision, and low latency.
-
Integrate human feedback loops to improve model performance.
-
Design scalable, observable, and robust ML systems.
-
Collaborate with researchers and SMEs to test edge cases and generate training data.
-
Work with engineers to integrate ML into production environments.
Requirements
-
3–8 years of experience building & deploying ML systems.
-
Strong foundation in traditional ML (clustering, anomaly detection, supervised learning).
-
Hands-on experience with LLMs (fine-tuning, embeddings, in-context learning).
-
Proficiency in Python, NLP tooling, and cloud platforms (AWS/GCP).
-
Experience with human-in-the-loop systems.
-
Skilled at balancing tradeoffs (recall, precision, latency, cost).
-
Strong communication skills across technical and non-technical audiences.
Nice to Have
-
Real-time ML pipelines or large-scale threat detection.
-
Multilingual embedding systems with code-switch detection.
-
Experience with OCR, vision, audio, or deepfake classification.
-
Familiarity with LangChain, Hugging Face, and error analysis/forensics.
-
Agentic pipelines for explainable or rationale-based moderation.
Success in the First 3 Months
-
Deploy a functioning moderation system using embeddings + fine-tuned classifiers.
-
Build an evaluation pipeline with precision/recall metrics and error analysis.
-
Contribute to data strategy (synthetic generation, clustering, or policy alignment).
-
Deliver a subsystem from ideation to deployment that holds up under real-world use.