I study electrical and computer engineering at Duke, build software for messy real-world problems, and go down a lot of rabbit holes about how the world actually runs.
hover to see connections, click to read more
What I've worked on, the questions I keep coming back to, the places in between, and what I wrote along the way. A line means a real connection, not a theme.
Places
Jobs
Projects
Questions
AaWriting
Places
Summer 2024, pitching LinkedIn Mentor at LinkedIn HQ.
Summer 2026, for Palantir.
Born and raised here until I was twelve. Back in 2025 for Fastbreak and the clinic project.
On the world mapCollege at Duke, from 2024, and the most fun place I've ever been: basketball, my frat, club hockey, endless memories. Also ECE, Duke Capital Partners and Ember.
On the world mapWhere my family's roots are. I taught English here with a local NGO the summer I was sixteen, and came back in 2024.
On the world mapI studied abroad here from August to the end of November 2026, and got to travel around Australia and Southeast Asia.
On the world mapJobs
Won their case competition in 2023, then came back in 2024 as their first intern, building LLM research agents that draft scientists' reports.
The summer I was seventeen, remotely: analysing latency and block propagation in the Ethereum infrastructure that trading firms run on.
Summer 2026 in New York as a deployment strategist intern: AI document workflows for an insurance customer, then procurement forecasting for a medical-supply one.
More on the workVoice and SMS agents for eight pediatric autism clinics in North Carolina, 2025. I led it and built the voice agent and most of the SMS side.
Read the case studySummer 2025 in Charlotte. A CMS for tournament operators and SAML single sign-on, shipped as an internal package.
Investment analyst since 2025. I diligence AI, SaaS and healthcare startups and write memos for the partners, which mostly means reading a lot of pitch decks.
Projects
An agent that matches students with mentors from their LinkedIn profiles. I presented it at LinkedIn HQ to co-founder Allen Blue in 2024.
A research tool that turns one question into a branching map. Third place at Neo Hackathon 2025, built by three of us.
How it worksA personal CRM that maps your network so you can find warm introductions. Won $1,500 from Duke Venture Group in 2026.
Try itA 2023 learning project: XGBoost against Prophet on hourly PJM data. XGBoost won, 9.2% error to 16.5%.
The write-upIn 2024 I worked on a small lease-to-own pilot in Tanzania: riders paying off their motorbikes through M-Pesa.
Questions
On the clinic project, the version that worked in a demo and the version I'd leave running at a real front desk were far apart. We started with several agents passing work between them and ended with one, plus plain code wherever a wrong answer mattered and a handoff to staff for anything unclear. I still don't have a clean rule for when an agent has earned less supervision.
Between a factory and a shelf, one container goes through trucking companies, two ports, a shipping line and customs brokers, and each keeps its own records. Some of that friction is physical: ships are slow and ports fill up. A lot of it seems to be email and PDFs. I'd like to know which part is which. It's a big reason Flexport interests me.
A lot of what I end up reading starts as an engineering story and turns into a political one: a subsea cable routed the long way around a contested sea, a shipping lane that empties out after a few attacks, export controls on chips. I'm not an expert in any of it. Mostly I'm trying to see where the engineering stops being the hard part.
In high school a mentor told me to pick one technical skill and become exceptional at it. I've come back to that a lot, partly because I've moved around: blockchain, forecasting, now agents. Running and hockey are where practice adding up is easiest to see. Whether the rest is reps, obsession, the people around you or some mix, I don't know yet.
AaWriting
2023. Why billions of people are still offline, and an argument for predicting bandwidth demand to squeeze more out of networks that already exist.
Read on MediumOne summer: LinkedIn HQ, interning at BenchSci, building Daylee, and the motorbike pilot in Tanzania.
Read on Substack2025, project lead and engineer
I spent weekends in North Carolina clinics watching how front-desk staff handled calls, texts, scheduling and insurance questions. Then I built voice and SMS agents to take the routine part of that off their plates. Anything the agent wasn't sure about went to a person. Once the system was stable, I handed it to an engineering team.
Summer 2026, New York
Deployment strategist intern
2023
Forecasting US electricity demand
XGBoost against Prophet on hourly PJM data. XGBoost won comfortably: 9.2% error against 16.5%. Write-up
2023
Forecasting bandwidth for smarter routing
Predicted internet bandwidth demand from ITU data, as a first step towards routing traffic better in under-served regions. Write-up
Everything else, including internships and the hackathon pile
A few questions I keep coming back to. Some come from things I've built; others are just rabbit holes I've fallen into. If you've thought about any of these, I'd love to talk.
When can you actually trust an AI agent?
On the clinic project, the version that worked in a demo and the version I'd leave running at a real front desk were far apart. We started with several agents passing work between them and ended with one, plus plain code wherever a wrong answer mattered and a handoff to staff for anything unclear. I still don't have a clean rule for when an agent has earned less supervision.
Can you engineer your way around geopolitics?
A lot of what I end up reading starts as an engineering story and turns into a political one: a subsea cable routed the long way around a contested sea, a shipping lane that empties out after a few attacks, export controls on chips. I'm not an expert in any of it. Mostly I'm trying to see where the engineering stops being the hard part.
Why is moving a container across the world still so complicated?
Between a factory and a shelf, one container goes through trucking companies, two ports, a shipping line and customs brokers, and each keeps its own records. Some of that friction is physical: ships are slow and ports fill up. A lot of it seems to be email and PDFs. I'd like to know which part is which. It's a big reason Flexport interests me.
How do you actually get exceptionally good at something?
In high school a mentor told me to pick one technical skill and become exceptional at it. I've come back to that a lot, partly because I've moved around: blockchain, forecasting, now agents. Running and hockey are where practice adding up is easiest to see. Whether the rest is reps, obsession, the people around you or some mix, I don't know yet.
August 2024, Substack
Pitching LinkedIn Mentor at LinkedIn HQ, interning at BenchSci, building Daylee, and running a small lease-to-own motorbike-taxi pilot in Tanzania paid through M-Pesa.
December 2023, Medium
XGBoost against Prophet on hourly PJM data, motivated by how much power new data centres need. XGBoost wins, 9.2% error to 16.5%.
October 2023, Medium
Why billions of people are still offline, how the internet's physical layer actually works, and an argument for squeezing more out of existing networks by predicting bandwidth demand.
All 20 pieces, mostly monthly updates from high school onwards
Seven places I've lived, worked or studied so far, from Charlotte to Sydney, each with a short story.