Applied AI Engineer, Agentic Analytics Platform

Overview:

The Business and Strategy Analytics (BSA) team is part of the Intelligence and Data Solutions function, partnering with the Asia Pacific Regional President’s Office to enable Visa’s internal business units to optimize performance and make intelligence-driven decisions. The team is at the forefront of transforming how intelligence is created and consumed across the organization — combining Visa’s proprietary data assets, frontier AI, and embedded human expertise to democratize trusted insight at scale.

Job Description:

This is an applied artificial-intelligence (AI) engineering role. You will help build and run the Agent for AP leadership, a full-stack analytical agent that lets users explore, analyze, visualize, and explain financial and transaction data in natural language.

Day to day, you will write Python and SQL, work in a Node.js and React codebase, design agent workflows and the tools they call, and be responsible for whether the answers the product gives are actually right. You will own defined workstreams end to end, with architecture and priorities set together with your team lead and with data, platform, security, and product partners contributing to the layers they own.

We expect strong fundamentals, real production experience in some of it, and the appetite to learn the rest.

Why this role:

MIS Agent is a live production system with real internal users, not a pilot. You will work on a modern agent stack that includes large language model (LLM) orchestration, a business knowledge layer, text-to-SQL, durable workflows, and an evaluation practice. You will also get end-to-end ownership that is unusual at this level: your work can go from a business question to a shipped, trusted answer.

Key Responsibilities:

Build the Analytical Agent - Primary Ownership:

Build and extend the agent harness, including goals, context, instructions, tools, validation checks, and escalation paths.

Design the workflow behind a user question, from data retrieval and SQL generation through validation, narrative, and visualization.

Make agent behavior reliable in production through structured outputs, tool calling, durable state, retries, timeouts, cancellation, human-in-the-loop controls, and explicit failure handling.

Apply context-engineering practices and help decide whether a problem warrants a chatbot, a fixed workflow, or an agent.

Make Its Answers Correct - Primary Ownership:

Build and maintain the knowledge layer that defines business terms, metrics, data structures, and expected answers.

Prepare trusted data for the agent to query, working with data teams to get grain, joins, and reporting logic right.

Validate agent output against trusted data sources and reference reports before release.

Build evaluation suites, benchmark datasets, and acceptance criteria, and use them as routine release gates.

Investigate discrepancies and recurring failure patterns, fix root causes, add regression coverage, and document known limitations, risks, and assumptions.

Apply appropriate analytical and statistical methods to investigate data, validate outputs, and explain findings; use machine-learning or distributed-processing methods where the problem genuinely requires them.

Ship and Operate It - Primary Ownership of Your Features; Hands-on Contribution Across the Shared Platform:

Develop and enhance the Python services, Node.js application programming interfaces (APIs), and React/TypeScript interfaces that make up the product.

Design asynchronous, stateful workflows that remain observable, testable, and maintainable in production.

Automate the path from data retrieval through analysis to publication.

Apply sound engineering practices through tests, documentation, code review, version control, instrumentation, containerized delivery, and production support.

Keep It Governed - Partnered with Platform, Security, and Data Teams:

Integrate with established enterprise identity and data platforms, and contribute to access-control, logging, rate-limiting, and error-handling behavior alongside the teams that own those controls.

Make sure the product uses approved data and communicates clearly when information is unavailable or not permitted.

Consider security, privacy, reliability, and responsible AI throughout development, and escalate concerns to the right owner.

Work with the People Who Use It - Shared Responsibility:

Gather requirements and translate them into explicit definitions, analytical logic, test cases, and working software.

Demonstrate the product, collect structured feedback, and surface barriers to trust and adoption.

Deliver practical guidance on prompting, AI workflows, and responsible use of generative AI.

Visa requires at least 3 days in office, expectations of these days will be confirmed by your Hiring Manager.

Qualifications

Basic Qualifications:

Bachelor's degree in a quantitative or technical discipline such as Computer Science, Software Engineering, Data Science, Information Systems, Statistics, Applied Mathematics, Operations Research, Engineering, or Analytics, or equivalent practical experience.

3-6 years of experience building analytical applications, data workflows, or AI-powered products, or equivalent demonstrated capability.

Demonstrated ability to independently deliver defined projects or technical workstreams.

Strong Python and SQL used for production code, including testing, logging, and debugging running applications.

Hands-on experience building LLM-powered applications, including instruction design, tool calling, structured outputs, context management, and explicit handling of failures and hallucinations.

Analytical rigor, including relational data modelling, reconciling results to trusted sources, investigating discrepancies, and turning ambiguous questions into definitions, test cases, and acceptance criteria.

JavaScript or TypeScript literacy: you can read, debug, and make scoped changes in an existing codebase and are willing to own a feature across an API and web in