Technical Program Manager II · Mastercard

Aditya Kulkarni

I take AI systems from architecture to production — and keep the engineering teams around them moving fast.

A TPM who operates closer to a tech lead than a coordinator. I read the system design doc, find the gap between what it promises and what the requirements demand, and get it resolved before the sprint starts.

Get in touch

St. Louis, MO 3+ yrs in program delivery

Aditya Kulkarni
3 AI systems shipped to production
35% Faster QA cycle time
40% Lower pipeline latency
6+ Cross-functional teams coordinated

Experience

Where I've shipped

Jul 2026 — Present O'Fallon, MO Current

Mastercard

Technical Program Manager II

  • Technical program management across the Services organization — driving delivery with engineering, product, and architecture teams on concurrent workstreams.
  • Partnering on platform architecture decisions and the Services delivery roadmap, translating between technical constraints and program commitments.
May 2025 — Jul 2026 St. Louis, MO

Spectrum

Technical Program Manager

  • Drove end-to-end delivery of 3 enterprise LLM systems — automated test case generation, code impact analysis, and SDLC workflow automation — from requirements intake through production across 4 engineering teams.
  • Co-designed the RAG pipeline architecture with engineers: Kafka ingestion → chunked embeddings on AWS Bedrock → pgvector store → FastAPI retrieval layer, cutting QA cycle time ~35%.
  • Built validation and observability workflows — schema checks, latency monitors, alert thresholds — reducing mean-time-to-detect on data issues by ~40%.
  • Coordinated 6 cross-functional teams (engineering, PMO, architecture, security, QA, vendors) across the full AI release lifecycle, from capability scoping through deployment and hypercare.
  • Eliminated ~30% of manual QA and documentation effort by embedding LLM-assisted workflows directly into developer tooling.
May — Jul 2024 St. Louis, MO

Spectrum

Technical Project Manager Intern

  • Ran Agile execution for 2 concurrent release workstreams — owned the Jira backlog, defined acceptance criteria for 40+ user stories, and drove bi-weekly releases with zero scope creep.
  • Restructured refinement sessions to align sprint goals to roadmap milestones, reducing carryover stories ~20% across an 8-week delivery window.
Jan 2023 — Dec 2024 Buffalo, NY

University at Buffalo

Research Assistant

  • Engineered a distributed Kafka message broker for IoT device fleets in Python — +30% throughput and −40% end-to-end latency versus a baseline polling architecture.
  • Built async WebSocket streaming infrastructure for high-throughput sensor telemetry, achieving sub-100ms response times at scale.

Selected work

Systems I've designed and delivered

Each of these started as a delivery problem, not a technology choice. The architecture followed from the constraint.

Feb 2026

Agentic AI Program Automation

Problem
Routine TPM work — research, scheduling, cost estimation — consumed 4–6 hrs/week of manual effort.
Built
Multi-agent system on AWS Bedrock where an orchestrator LLM routes tasks to specialist agents via tool-use APIs.
Tradeoff
Tool-use over fine-tuning — full flexibility, zero retraining overhead.
Impact
~60% reduction in time-on-task in prototype testing.
PythonLLM AgentsAWS BedrockTool Use
Dec 2025

Real-Time Risk Intelligence Engine

Problem
Program blockers only surfaced in status meetings — 2–3 days after the signal already existed.
Built
Kafka pipeline → Isolation Forest anomaly detector → FastAPI risk-scoring API → Grafana dashboard with per-workstream indicators.
Impact
Warnings surfaced 48–72 hrs ahead of status reporting; eliminated surprise blockers across 10+ workstreams.
Kafkascikit-learnFastAPIGrafana
Nov 2025

AI-Powered SDLC Intelligence Platform

Problem
QA teams spent 40%+ of sprint time hand-writing test cases and mapping code-change blast radius.
Built
Commits → FastAPI ingestion → chunked embeddings → pgvector → Bedrock retrieval → structured test case and impact report.
Tradeoff
RAG over fine-tuning — lower cost, no stale-model problem, the codebase stays live.
Impact
~35% less QA cycle time; test case generation fell from ~2 hrs to ~15 min per sprint.
LangChainRAGpgvectorFastAPI
Jan 2025

Delivery Analytics & Capacity Dashboard

Problem
Engineering teams had no visibility into delivery trends or capacity gaps — planning was reactive and spreadsheet-driven.
Built
Python/SQL analytics pipeline feeding 6 Tableau dashboards with rolling 6–8 week forecasting models.
Impact
Replaced manual planning for 3 teams; surfaced utilization bottlenecks before they hit sprint commitments.
PythonSQLTableauForecasting

Approach

How I operate

My research background is in distributed systems and real-time data pipelines. That means I can read a design doc, spot the gaps, and align engineers on tradeoffs without needing anything translated first.

01

Ambiguity

I decompose unclear problems into the smallest testable assumption and work forward from there — not from consensus, and not from the loudest voice in the room.

02

Technical systems

I read architecture docs the same way I read PRDs: looking for the gap between what the system promises and what the requirements demand. I flag those before the sprint starts.

03

Cross-team execution

Coordination failures are communication failures. I solve them with shared artifacts — decision logs, dependency maps, clear owners — not more meetings.

04

AI & LLM workflows

I evaluate AI systems by their failure modes first — hallucination rate, retrieval recall, latency tail cases — before celebrating accuracy on the happy path. Production is not a demo.

05

Delivery risk

I treat delivery risk the way engineers treat technical debt: track it, quantify it, and surface it early — before it compounds into a missed release.

06

Ownership

I don't hand a blocker to someone else and call it managed. If it sits on the critical path, it's mine until it's resolved or genuinely owned by a named person with a date.

Toolkit

What I work with

Program delivery

End-to-end executionMulti-workstream deliveryRisk mitigationDependency mappingCapacity planningRoadmappingStakeholder comms

Agile & process

Agile / ScrumSAFeJiraConfluenceRelease managementBacklog prioritization

AI & data

RAG pipelinesLLM agentsAWS BedrockPythonSQLTableauA/B testing

Platform & infrastructure

KafkaKubernetesPostgreSQL / pgvectorFastAPIJenkinsGitLab CIDistributed systems

Credentials

Certifications & education

Professional certifications

Project Management Professional Project Management Institute
Certified SAFe 6 Agilist Scaled Agile, Inc.
AWS Certified Cloud Practitioner Amazon Web Services

Coursework

Deep Learning Specialization · DeepLearning.AI Google Project Management · Coursera AI for Everyone · DeepLearning.AI

Education

B.S. Computer Science & Engineering

University at Buffalo, State University of New York

Graduated May 2025 · Buffalo, NY

Contact

Let's talk delivery

Always up for talking technical program management, AI delivery, and hard systems problems — comparing notes, trading war stories, or just connecting. I work best where the problems are hard, the systems are real, and execution actually matters.

Email adityamk22@gmail.com
LinkedIn linkedin.com/in/adityakulkarnii
Phone 631-633-8025