Skip to content
BMBrian Maina

Computer Science + Economics / Boston University / Class of 2028

Brian Maina

Most of my work is backend systems: APIs, caching layers, data pipelines, and a C++ matching engine. I'm looking for Summer 2027 SWE internships in fintech and backend infrastructure.

C++ / Python / TypeScript / FastAPI / PostgreSQL / Redis / Docker

Now

  • Barnes GroupUndergraduate Research Assistant (REU)
  • Marintchev LabUndergraduate Research Assistant
  • Hack4ImpactJunior Developer
  • Alpha Kappa PsiVice President of Administrative Affairs
  • Summer 2027Seeking SWE internship
See projectsRésumé
Brian Maina

Boston, MA. Currently building backend systems in Python and C++.

By the numbers

250K
orders/sec sustained
89µs
p50 order-to-ack
1,000+
CRM profiles synced
32×
faster stop lookups

About

I'm a Computer Science and Economics student at Boston University, graduating in 2028. Most of what I build is backend: APIs, caching layers, data pipelines. Lately I've been going deeper on systems programming in C++, and this fall I joined the Barnes Group at BU to research agentic AI for science.

This summer at AdaptX I wrote the nightly sync that pulls donors, runners and grant applicants out of Zeffy, RunSignUp and JotForm into one CRM. Three platforms describe the same person three different ways, and deciding when two records are one person turned out to be the hard part. That kind of problem, where the data has to be right, is what I want to keep working on.

Currently: Agentic AI research, Barnes Group / NMR assignment pipeline, Marintchev Lab / C++ systems programming

What I'm into

Backend

FastAPI and Node services, PostgreSQL, Redis and Docker, deployed on Render, Railway, Vercel and DigitalOcean. Plus the scheduled jobs that keep data in sync.

Fintech & markets

An exchange matching engine, a replication of a market-microstructure paper on NASDAQ data, and parsing bank statement exports that all disagree with each other.

Systems programming

C++17: a TCP order gateway, a single-writer matching thread, and a write-ahead log that survives kill -9. Tested under ASan, UBSan and ThreadSanitizer.

Tools I've used

Languages

Python, C++, JavaScript, TypeScript, SQL, Swift

Frameworks & libraries

FastAPI, SQLAlchemy, React, React Native, Node.js, Express, SUMO

Databases & infrastructure

PostgreSQL, Redis, Docker, Supabase, Firebase, DigitalOcean

Tools & platforms

Git, GitHub Actions, Jest, Supertest, pytest, Vercel, Railway, Render

Education

B.A. Computer Science and Economics

Boston University / Boston, MA / Expected May 2028

Coursework

  • Data Structures & Algorithms
  • Computer Systems
  • Software Engineering
  • Databases
  • Linear Algebra
  • Theory of Computing Languages
  • Empirical Economics
  • Economics in Legal Issues
  • Discrete Math

Projects

Featured / DivHacks 2026 at Columbia UniversityWinner, MLH Best Use of DigitalOcean · DivHacks 2026

Ellipsis

Two-person hackathon team

Detect a blockage.Measure the impact.Test a response.

Built for the operators at NYC DOT's traffic management center. A blocked lane on one of 10 live DOT cameras opens an incident with the camera evidence. SUMO then runs the current signal timing against a bounded retiming, and the operator sees both before deciding. Simulation is decision support; a person still decides.

SUMO / Python / FastAPI / React / MapLibre / Docker / DigitalOcean

What I built

  • Built the SUMO model of Midtown: 88 signalized intersections, used to score each bounded retiming plan against the current timing.
  • Built the operator dashboard in React and MapLibre: incidents by severity, the camera evidence, and a base-versus-simulation comparison before Accept or Reject.
  • Ran it live on NYC DOT camera feeds, deployed on DigitalOcean with Docker Compose. The public site falls back to a recorded replay if the server is down.
  • Detection and tracking were my teammate's. On 15 hand-tagged blockages the team measured 88% precision and 93% recall.
88
signalized intersections in SUMO
10
live NYC DOT cameras
63s
median time to alert (team)
2 days
two-person build
Project 01In progress
order-matching-engine / README.md
## Measured results ack p50     45 µs @ 5K/sp50/p99     89/149 µs @ 50K/ssustained   250K orders/sfsync       +11 µs p50kill -9     50/50 recoveredtests       191, ASan UBSan TSan

Order Matching Engine

A limit order book in C++17 behind a TCP gateway, with a write-ahead log so the book survives being killed. My first C++ project.

  • One matching thread owns the book and gets orders over a lock-free SPSC queue, so no lock sits near it. 89µs p50 order-to-ack at 50K orders/sec, 250K/sec sustained.
  • Recovered the book in 50 of 50 randomized kill -9 runs, checked against an independent reimplementation. 191 tests under ASan, UBSan and TSan.
Year
2026–
Lang
C++17
Scale
250K

250K orders / sec

Open →
Project 02In progress
github.com / pulls
  • nesquena/hermes-webui #5915Merged

    fix(#4732): keep mobile composer-mic dictation active through pauses (salvage of #5294)

  • apache/superset #41587Merged

    fix(i18n): update German security menu translations

Open Source

Pull requests to projects I actually use.

  • Hermes-WebUI (18k★): fixed mobile dictation cutting out at the first pause. The maintainer rebuilt my PR to avoid a desktop regression, shipped it, and credited me as co-author.
  • Apache Superset (74k★): merged a fix correcting the German translations in the security menu.
Year
2026–
Lang
JavaScript
Scale
18k★

18k★ Hermes-WebUI

Open →
Project 03Complete
TriLive for iOS
TriLive app icon

TriLive

Team project

A real-time Portland transit tracker for iOS that I co-built. It reached 70+ users on the App Store and is no longer listed. I owned the backend and the caching layer.

  • Built the FastAPI backend and the TriMet ingestion layer that serves live vehicle positions.
  • Got stop lookups from 16ms to 0.5ms with Redis caching and Geo queries for nearby stops. That was the first time caching made an obvious difference for me.
Year
2025–2026
Lang
FastAPI
Scale
32×

32× faster lookups

Open →
Project 04Complete
statement-wrapped.vercel.app
StatementWrapped spending breakdown interface

StatementWrapped

A tool that takes raw bank statement exports and turns them into a readable breakdown of where the money actually went. No login and no bank connection.

  • Normalized CSVs from Chase, Bank of America, Wells Fargo and Apple Card, which all disagree about column names, date formats and amount signs. Under 1ms per 100 rows.
  • Categorized transactions in one batched LLM call per statement, falling back to "other" when the model fails so ingest never crashes.
Year
2026
Lang
FastAPI
Scale
4

4 bank formats

Open →
Project 05In progress
project-winston / backend / app
ml/  regime.py         # HMM regimes  trainer.py        # training  stocks_ranker.py  # LightGBM  sizer.py          # Kelly sizingdata/  fetcher.py  eia_fetcher.pyagents/  overseer.py  pipeline.py

Project Winston

A research platform for market data. Ingestion, feature engineering and model serving, across 17 commodity futures and 503 S&P 500 names.

  • Built the ingestion and feature pipeline, then stacked XGBoost and LightGBM classifiers on top with walk-forward out-of-sample validation.
  • Added HMM regime detection to gate signals, so they get suppressed in conditions where the model has not been reliable.
Year
2026–
Lang
Python
Scale
17

17 commodity futures

Open →
Archive5 projects

Older and smaller builds

Where I've worked

  1. Sep 2026 – Present

    Undergraduate Research Assistant (REU)

    Barnes Group / BU Faculty of Computing & Data Sciences / Boston, MA

    • Researching agentic AI for science in Prof. Elizabeth Barnes's AI for weather and climate group.

    Python / BU Shared Computing Cluster

  2. May 2026 – Aug 2026

    Software Engineer Intern

    AdaptX / Nonprofit, fitness accessibility platform / Boston, MA

    • Built the React Native and Firebase app for adaptive athletes, with Cloud Functions in TypeScript and Firestore: virtual 5K participation, race registration and fundraising.
    • Wrote the nightly Python sync that pulls donors, runners and grant applicants from Zeffy, RunSignUp and JotForm into Bloomerang CRM, with fuzzy-match dedup. Live runs merged 1,000+ profiles and 72 campaigns.
    • Automated ingestion of 13K race photos, and built an admin dashboard for moderating community posts and training guides.
    • Migrated the app's auth from Clerk to Firebase Auth, and added Jest suites and Maestro end-to-end flows.
    ZEFFYRUNSIGNUPJOTFORMDIFFERENT SCHEMASNIGHTLY, GITHUB ACTIONSPYTHON SYNCNORMALIZE + DEDUPBLOOMERANGCRM
    Three platforms describing the same people differently, merged into one CRM every night.

    React Native / Expo / Firebase / TypeScript / Python / GitHub Actions / Jest / Maestro

  3. Mar 2026 – Present

    Undergraduate Research Assistant

    Marintchev Lab / BU Dept. of Pharmacology, Physiology & Biophysics / Boston, MA

    • Contributing to a Python pipeline that automates NMR backbone assignment for protein 5MP1, toward a lab publication.
    • Improved carbon peak matching accuracy 53% by cross-referencing backbone (Cα/Cβ) and side-chain shift lists.
    • Added proton-shift filtering to the pipeline, to recover sequence information from lower-quality spectra.

    Python / Jupyter / NMR / CARA

Clubs & leadership

  1. May 2026 – Present

    Vice President of Administrative Affairs

    Alpha Kappa Psi / Nu Chapter, Boston University / Boston, MA

    • Built and still maintain the chapter member portal in React, Node/Express and PostgreSQL. 70+ members use it.
    • Added a family-tree visualizer, alumni directory search and a life-points engine, to replace a pile of spreadsheets.

    React / Node.js / Express / PostgreSQL

    GitHubLive
  2. May 2026 – Present

    Junior Developer

    Hack4Impact / Boston University / Boston, MA

    • Building front-end features for the Boston Municipal Research Bureau site redesign, which city policymakers use.
    • Moving a legacy WordPress site onto something accessible and searchable, covering decades of research.

    Next.js / React / Accessibility

  3. May 2025 – May 2026

    Treasurer

    German Culture Club / Boston University / Boston, MA

    • Managed a $1,780 semester budget across events, reimbursements and purchase requests.
    • Membership grew 55% while I was treasurer, from 60 to 93.
    • Kept a ~$1,000 reserve fund to stabilize operations across funding cycles.
  4. Sep 2024 – May 2025

    Assistant Outreach Lead

    Engineers Without Borders / Boston University chapter / Boston, MA

    • Brought projected well-construction costs down 60%, from $50,000 to $20,000, by finding alternative contractors.
    • Cut turnaround 30% by coordinating 5+ engineers on bid analysis for the Ogiek Kwanza Project.
  5. Jun 2023 – Sep 2023

    Co-founder

    Digify Dubai / Digital literacy initiative / Dubai, UAE

    • Co-founded a digital literacy initiative that trained 30+ migrant workers a week.
    • Taught sessions on language learning and mobile banking.
    • Sourced and set up 20 laptops so every session had enough machines to run.

    Featured in Khaleej Times

GitHub activity

Contributions

LessMore

2,070

Last year

5 days

Current

12 days

Longest

26

Repos

Recent commits

Get in touch

Email is the fastest way to reach me.