Bank of America · Global Technology · Summer 2026
Enterprise AI, grounded in how people actually work.
Business Analyst Intern · AI Strategy & Product Discovery
Across three AI initiatives, I moved from user problems to product direction: interviewing Business Analysts, designing context-aware workflows, analyzing an undocumented mainframe system, and prototyping employee onboarding support.
Mosaic / simplified system view
Context before generation
context engine
Mosaic
“The problem wasn’t access to AI. It was missing workflow context.”
15
BA interviews
600+
legacy files
03
AI initiatives
01
internal review
my scope
My formal title was Business Analyst, but the primary assignment functioned like early product work: research the workflow, define the problem, shape the direction, and help build the solution.
01 / product discovery
Mosaic
With two fellow BA interns, I conducted 15 semi-structured interviews across two offices to understand why existing generative tools were not reducing Business Analyst workload.
what we heard
BAs repeatedly drafted epics, user stories, acceptance criteria, documentation, and Jira handoffs—then spent additional time correcting generic AI output.
what it meant
The tools lacked project-specific context: codebase knowledge, historical stories, team language, and the connective tissue between documentation and delivery.
We stopped asking, “What should the chatbot write?” and started asking, “What context does the workflow need?”
product direction
A context-aware agent system
simplified · confidential details omitted
context
Knowledge base
handoff
Jira-ready work
recognition
Submitted for internal patent review
I contributed to Mosaic’s architecture, workflow design, and supporting documentation.
enablement
Weekly BA office hours
I trained active BAs on prompt structure, workflow integration, and tool navigation.
02 / legacy modernization
5500 Documentation Agent
I was assigned to a 401(k) mainframe system containing 600+ COBOL and VSAM files with almost no active documentation beyond a 2010 slide deck—and no prior COBOL background.
the question
How do you make an undocumented system understandable without manually decoding every file?
input
600+ files
COBOL + VSAM
agent
5500 engine
parse + synthesize
output
system map
logic + dependencies
System logic
A high-level view of modules and responsibilities.
File relationships
Dependencies and data movement across the codebase.
Reusable model
A generalized approach for other legacy applications.
03 / cross-functional codeathon
AI Onboarding Agent
Working with software engineers, Business Analysts, and finance teammates, I designed the UX flow and conversational persona for a Copilot Studio assistant that delivered team-specific context to new hires.
Team context
Role-specific codebase and workflow guidance.
Guided Q&A
Answers for recurring setup and process questions.
Less repetition
Reduced intended support burden for senior teammates.
handoff + impact
The work continued after the internship.
Patent review
Mosaic architecture and documentation entered the internal review process.
Reusable model
The documentation approach was generalized beyond one codebase.
Return offer
Received a full-time Corporate Technology return offer.
what I took with me
Enterprise AI succeeds when user research informs the system architecture—not when another chatbot is added to the workflow.