Machine Learning Engineer, Abnormal Security

• 3+ years experience designing, building and deploying machine learning applications in one of the domains of text understanding, entity recognition, NLP experience, computer vision, recommendation systems, or search
• 1+ years of experience with writing stable and production level pipelines for model training and evaluation leading to reproducible models and metrics
• Experience with data analytics and wielding SQL+pandas+spark framework to both build data and metric generation pipelines, and answer critical questions about system efficacy or counterfactual treatments
• Ability to understand business requirements thoroughly and bias toward designing a simplest yet generalizable ML model / system that can accomplish the goal
• Uses a systematic approach to debug both data and system issues within ML / heuristics models
• Fluent with Python and machine learning toolkits like numpy, sklearn, pytorch and tensorflow
• Effective software engineering skills who can find answers quickly from code base and writes structured, readable, well tested and efficient code
• BS degree in Computer Science, Applied Sciences, Information Systems or other related engineering field
• This position is not:
• A role focused on optimizing existing machine learning models
• A research-oriented role that's two-steps removed from the product or customer
• A statistics/data science meets ML role

Desirable

• Abnormal Security is looking for a Machine Learning Engineer to join the Message Detection - Attack Detection team
• At Abnormal, we protect our customers against nefarious adversaries who are constantly evolving their techniques and tactics to outwit and undermine the traditional approaches to Security
• This team is solving a multi-layered detection problem, which involves modeling communication patterns to establish enterprise-wide baselines, incorporating these patterns as robust signals, and combining these signals with contextual information to create extremely precise systems
• The team builds discriminative signals at various levels including message level (eg. presence of particular phrases), sender-level (eg.frequency of sender) and recipient level (eg.likelihood of receiving a safe message)
• These signals are then combined and utilized to train highly accurate model based as well as heuristic detectors
• Additionally, to continuously adapt to new unseen attacks, the team builds out different stages in our automated model retraining pipelines including data analytics and generation stages, modeling stages, production evaluation stages as well as automated deployment stages
• This role would also have an opportunity to have a significant impact on the overall charter, direction and roadmap of the team
• The Machine Learning Engineer would be involved in understanding the domain of false negatives i.e. the current and future attacks which can cause significant customer workflow disruption
• They would help define the technical roadmap required to address the most pressing customer problems and simultaneously operate our detection decisioning system at an extremely high recall
• Design and implement systems that combine rules, models, feature engineering, and business and product inputs into an email detection product, with senior engineer guidance
• Understand features that distinguish safe emails from email attacks, and how our model stack enables us to catch them
• Identify and recommend new features groups or ML model approaches that can significantly improve detection efficacy for a product. Work with infrastructure & systems engineers to productionize signals to feed into the detection system
• Writes code with testability, readability, edge cases, and errors in mind
• Train models on well-defined datasets to improve model efficacy on specialized attacks
• Actively monitor and improve FN rates and efficacy rates for our message detection product attack categories, through feature engineering, rules and ML modeling
• Analyze FN and FP datasets to categorize capability gaps and recommend short term feature and rule ideas to improve our detection efficacy
• Contribute in other areas of the stack: building and debugging data pipelines, or presenting results back to customers in our tools when the occasion arises