COMPANY INFORMATION:

Our client is an AI-powered enterprise automation company that helps large organizations streamline complex operational workflows through intelligent AI agents. Its technology captures how employees perform tasks across multiple applications, transforms those activities into repeatable workflows, and enables AI agents to execute them autonomously, with human oversight and approvals when needed.

Our client serves major enterprise customers, including U.S. financial institutions and a leading managed-cloud services provider.

JOB OVERVIEW:

Automate real business workflows by turning desktop recordings and system logs into agent-driven processes that pass rigorous evals. Our Client builds applied AI for business process automation, mixing rules, classical ML, and LLMs to get reliable outcomes at sane cost and latency. Work covers workflow discovery from recordings, outcome definition, and end-to-end eval suites across recorded and synthetic sessions. Day to day is Python and SQL for data pipelines, model selection and tuning, and agent construction that stands up in production.

Our Client is hiring mid to senior applied engineers who ship to production and work directly with customer teams in operations, IT, and the business. The role is US-based, fully remote, and customer-facing. Delivery happens inside each customer’s cloud with privacy designed in from the start. Day to day, you pick the method that fits the task, run tight evaluations with checks to catch regressions, and keep agents in your own dev loop. If measurable outcomes and automations that non-engineers actually use matter to you, this is worth a close look.

JOB RESPONSIBILITIES:

• Sit with operations, IT, and business stakeholders to map problems, run discovery sessions, and stay focused on outcomes
• From screen and audio captures, map a workday into discrete tasks, spot recurring processes across people and systems, and document the steps and the rules behind them
• Define clear workflow outcomes, then write evals that verify completion
• Create and run evaluation suites from both real and synthetic sessions, define ground truth and labels, track metrics, and add regression checks to every change
• Own the data flow from raw capture events and screenshots through timelines, engineered features, and labeled examples
• Privacy-by-design handling, including removing personal data and keeping all customer data in the customer's environment
• Method selection per step, choosing among rules, classical ML, small or fine-tuned models, and frontier models to balance accuracy, cost, and latency
• Design agent-driven workflows that produce automations, validate them against eval results, and route to a human for approval when needed
• Treat on-screen text as untrusted input so documents or web pages in a session cannot steer the model

REPRESENTATIVE PROJECTS:

• Process Discovery Benchmarking: Develop a labeled dataset of recorded employee workflows, identifying the start and end points of each process, and establish measurable benchmarks for discovery accuracy.
• Workflow Classification and Matching: Build a classification model that identifies when different recorded activities represent the same underlying business process, with measurable accuracy and reliability.
• Automated Workflow Outcome Validation: Implement a validation framework that compares AI-executed workflow results against human-performed outcomes to ensure accuracy and consistency for a selected customer process.
• AI Model Performance Optimization: Evaluate frontier AI models against smaller or fine-tuned alternatives for workflow discovery, benchmarking accuracy, execution cost, and response latency to identify the most efficient solution.
• Continuous Regression Testing: Develop an automated regression testing suite that evaluates every change to the workflow discovery engine, detects performance degradation, and prevents updates that negatively impact accuracy or reliability.

JOB REQUIREMENTS:

• 5+ years building ML or LLM systems in production where outcomes depend on users
• End-to-end eval design experience, including dataset creation, labeling, metrics, and using results to influence decisions
• Built pipelines for messy event and log data, including sequences, sessions, and time series, using Python and SQL
• Hands-on with ML methods such as classification, clustering, and sequence segmentation
• History of picking a smaller model over a frontier model when the metrics justify it
• Built agentic systems with tool use and quantified where they fail
• Clear writing that explains results and their limits
• Up-to-date on agent capabilities and failure modes today, with an opinion on how they will change over the next year
• Deployed an agent to a non-engineer business process such as ops, support, finance, or back office
• Can articulate hurdles like approvals, accountability, audit, permissions, legacy or no-API systems, and when a human-in-the-loop is necessary
• Not a fit: research-only without production, prompt-only roles, or demo-only agent experience
• Process mining, task mining, or RPA discovery is a plus
• Experience with computer-use agents and benchmarks such as OSWorld is a plus
• Vision-language work on screenshots or UI surfaces is a plus
• Fine-tuning and serving smaller models, including deployment in a customer cloud such as Azure OpenAI, is a plus
• Detecting and removing personal data in text and images is a plus