Senior ML / AI Engineer

What matters most

• Own production ML systems

  Lead work from research and prototyping through deployment and monitoring, then integrate AI capabilities into Confido’s platform with engineering and product.

• Extract financial-document data

  Build LLM or NLP systems that turn complex invoices, deductions, and retailer reports into structured data.

• Retrieve and reason across enterprise systems

  Develop agentic workflows that use information spread across fragmented or legacy systems.

• Forecast and detect anomalies

  Build models for brand sales and financial performance forecasts and for anomaly detection in retailer and sales data.

• Optimize retailer promotions

  Develop recommendation systems that help brands make decisions about retailer promotions.

Likely challenges to think through

• Accuracy versus coverage in document ingestion

  A likely design challenge is handling varied financial documents without silently extracting incorrect values. Prepare to discuss validation, uncertainty handling, and how you would measure extraction quality; this is a preparation hypothesis, not a stated Confido problem.

• Grounding workflows in fragmented data

  Retrieval across legacy systems may trade breadth of access against trustworthy, traceable answers. Prepare an approach to source attribution, missing information, and failure handling; the listing does not specify the systems or their condition.

• Useful predictions versus misleading signals

  Forecasts, anomaly alerts, and promotion recommendations may involve different costs for false alarms and missed opportunities. Prepare to explain evaluation choices and how you would connect model outputs to product decisions; these trade-offs are inferred from the stated duties.

Topics to prepare

• LLM and NLP document understanding

  Review how to extract structured fields from invoices, deductions, and retailer reports, including how to handle ambiguous or missing values. Exercise: sketch a field schema and define checks for extraction errors. Confido explicitly calls for financial-document extraction and experience with LLMs or NLP.

• Production ML lifecycle

  Review the decisions involved in moving from research and prototypes to deployment and monitoring. Exercise: outline what you would test before releasing a document-extraction model and what failures you would monitor afterward. The posting assigns ownership across this lifecycle and requires production ML experience.

• Retrieval and agentic reasoning

  Review how a workflow can retrieve information from fragmented or legacy systems, reason over conflicting results, and surface uncertainty. Exercise: diagram a retrieval-to-answer flow with failure cases. This is listed as a responsibility; prior agentic-workflow or retrieval-system experience is nice to have.

• Forecasting and anomaly detection

  Review time-aware evaluation for sales and financial forecasts and ways to distinguish unusual retailer data from expected variation. Exercise: propose a validation plan for a brand-sales forecast and an anomaly alert. Both are explicit example problems.

• Promotion recommendation objectives

  Review how a recommendation objective, constraints, and evaluation measure affect retailer-promotion decisions. Exercise: state what a promotion model would optimize and how you would check whether its recommendations help. Promotion optimization is an explicit responsibility; the specific objective is a preparation choice, not a stated requirement.

• Product-minded ML tradeoffs

  Review how to weigh model quality against integration effort and operational reliability. Exercise: compare two approaches to automating invoice ingestion and explain which outcome matters to a product partner. The posting requires a strong product mindset and collaboration with engineering and product.

Skills to revise

• Implement a document-extraction prototype

  Practice turning sample invoice or deduction text into validated structured records, then inspect incorrect and missing fields. This mirrors the posting’s financial-document ingestion work; no particular tool is specified.

• Design a deployable ML service

  Practice drawing the inputs, prediction path, integration boundary, monitoring signals, and fallback for one Confido example problem. This prepares for the stated production-ML responsibility and the listed ML systems design interview session.

• Build a small retrieval workflow

  As a suggested exercise, combine a few fragmented sample records, retrieve evidence for a question, and make the output traceable to those records. This reflects the agent-workflow responsibility, without treating any particular architecture as required.

• Evaluate forecasts and anomaly flags

  Practice making time-ordered train/test splits on sample sales data, comparing forecasts with a simple baseline, and reviewing false anomaly alerts. These exercises map to the stated forecasting and retailer-data anomaly tasks.

• Prototype promotion ranking

  As a suggested exercise, rank sample retailer promotions against an explicitly chosen objective and explain the tradeoffs to a product partner. Confido calls for promotion recommendation systems and practical applications, but does not prescribe a modeling method.

• Prepare a production ML project walkthrough

  Choose a real project you can discuss. Prepare a 90-minute walkthrough covering your decisions, LLM or NLP work if applicable, deployment, monitoring, results, and collaboration with engineering and product. The posting lists a 90-minute on-site project session.

• Sketch a financial-document extraction system

  For a practice design exercise, outline how you would turn invoices, deductions, or retailer reports into structured data. Include evaluation, handling of uncertain fields, and how outputs would reach the platform; these are preparation choices, not a stated interview assignment.

• Design a retrieval workflow for fragmented systems

  Draw a possible agentic workflow for retrieving information from legacy systems and reasoning over it. Be ready to explain how you would check retrieved evidence and handle missing or conflicting records.

• Work through a CPG modeling example

  Prepare a small example for forecasting brand sales or financial performance, detecting retailer-data anomalies, or optimizing promotions. State the target, baseline, evaluation approach, and how a brand would use the result.

• Document lifecycle and product tradeoffs

  Make a one-page account of how you have moved an AI/ML capability from prototype to production and monitoring, where your experience supports it. Identify decisions made with product and engineering and what you would improve.

• Rehearse the advertised interview formats

  Practice a timed coding problem and a 60-minute ML systems design discussion. The posting also lists a 20-minute recruiting call, a 60-minute technical screen, and an on-site session with coding, project, and systems design; it does not specify their exact questions.

• Bring focused questions for Confido

  Write questions about document-extraction quality, access to fragmented enterprise data, forecasting and promotion objectives, and how engineering and product evaluate deployed models. Prioritize what you need clarified rather than assuming current processes.