VT/

Senior Software Engineer

Vishal Tak

I build the backend systems that move loan decisions, approvals, and compensation through Capital One at scale and lately, the AI tooling that speeds up how my team ships them.

Based in
Frisco, TX
Focus
Distributed systems, applied AI
Currently building
Approval and workflow platforms

Selected Work

Approval Workflow Platform

2026

Problem: approvals across sales teams were scattered through email and Slack threads, with no audit trail or status visibility.

My role: architected and built the event-driven approval microservice end to end. Schema, state machine, and rollout.

Outcome: one system now handles 100+ approvals a day with a full audit trail and live status tracking.

Spring BootPostgreSQLAWS FargateState Machines
300+

sales agents running approvals through a single auditable system

AI-Driven Observability Workflow

2026

Problem: standing up dashboards and alerts for a new service took engineers days of manual configuration.

My role: designed and built a tool that generates dashboards and alerts directly from a service’s metrics on request.

Outcome: setup dropped from days to minutes, and the project took Runner-Up at a company-wide hackathon.

LLM ToolingNew Relic APIPython
DaysMinutes

to stand up dashboards and alerts for a new service

Loan Decisioning Graph Engine

2022-2024

Problem: the core decisioning engine ran single-threaded, making it the bottleneck as application volume grew.

My role: refactored execution into a multi-threaded worker pool, a 50% latency cut, and built the graph-authoring UI engineers still use today.

Outcome: the platform orchestrates ~1,000-node DAGs across six microservices with zero execution failures.

JavaAngularGremlinConcurrency
80K+

loan applications decisioned per day on the platform

About

I’m currently a software engineer at Capital One, working on the systems behind auto lending. The services that decide loans, route approvals, and calculate dealer compensation. Most of it lives in the part nobody sees: making sure work happens in the right order, holds up under volume, and fails in ways someone can actually debug at 2am.

The parts I enjoy most are the unglamorous ones. Clear ownership, useful logging, tests that catch real problems. Lately I’ve been looking for places where AI can take the repetitive work off my team’s plate, which is how the observability tooling above came about.

Outside of work you’ll usually find me on a basketball or pickleball court, at a piano, or planning the next trip.

Education
B.S. Computer Science, UT Austin
Certification
AWS Solutions Architect, Associate

Let’s build
something.