$ whoami

Francis
O’Hara Aidoo▍

I enjoy writing code to solve challenging problems, from frontend and backend to cloud infrastructure.

CS @ Colby College '27 · SWE Intern @ Google ('24, '25) · Platform AI @ Medidata ('26) · Co-founder @ StyleSyncs

Download Resume View Work

press ⌘K to navigate

~/avatar.png

now: open to 2027 SWE roles

$19.2M

ARR opportunity unblocked at Google

Officially published Terraform resource I built for Vertex AI

100+

users & $9K raised

StyleSyncs, the startup I co‑founded

3B+

CSV records/hour throughput

Apache Beam transform I built for Google Cloud Dataflow

70%+

higher table capture rate

Vision LLM pipeline I built at Medidata, tested on 45 protocol documents

10+

CS courses TA'd

Including Data Structures and Neural Networks at Colby

00 · cat about.md

I'm a Computer Science student at Colby College with a concentration in AI, graduating in May 2027.

My first exposure to programming was in fifth grade in Ghana, when my older brother, who was studying computer science at university, noticed I was curious about the projects he was working on. When he wasn't using his laptop, he let me borrow it to watch Udacity's Intro to Computer Science course, which taught Python by building a basic search engine. I found it challenging, but from there I fell in love with programming and kept coming back to it.

Then in 2019, I joined a computer science bootcamp for high school students run by ELiTE Education and started building projects with Python, Arduino, and Raspberry Pi.

I've since interned at Google twice, on the Vertex AI and Apache Beam teams. This past summer I was a data science intern on the Platform AI team at Medidata. I also co‑founded StyleSyncs, a virtual try-on startup, where I build the AI try-on features and the web app, and am now building our iOS app.

What motivates me most is building things people actually use.

01 · experience.log

Experience

My experience spans two software engineering internships at Google, a data science internship at Medidata, a startup I co-founded, and teaching assistant work at Colby College.

Medidata Solutions
Data Science Intern · Platform AI
May 2026 → Aug 2026 · New York, NY
Built LLM pipelines with frontier models on Amazon Bedrock for Medidata's clinical trial platform, which powers 37,000+ trials worldwide.
- Built a vision-based LLM pipeline in Python for Medidata's Protocol Digitizer service that extracts structured data from clinical trial protocol documents. It captured the Schedule of Assessments tables in 45 of 45 evaluation documents, compared with about 26 of 45 for the existing text-based workflow.
- Improved the reliability of the evaluation pipeline for eConsent, an LLM service that turns consent form PDFs into fillable e-forms. I analyzed 500 historical forms with Python, Pandas, Matplotlib, and Jupyter and found 6 field types it wasn't scoring.
- Extended eConsent's evaluation pipeline with per-language and per-page accuracy metrics and an HTML report with Plotly charts. I also implemented its migration to a newer Claude model on Amazon Bedrock ahead of the old model's end of life.
PythonAmazon BedrockClaudeLLMOpsPandasAmazon S3
Google
Software Engineering Intern · Vertex AI
May 2025 → Aug 2025 · Sunnyvale, CA
Built public infrastructure for deploying generative AI models at scale on Google Cloud.
- Built an officially published Terraform resource in Go for deploying generative AI models on Vertex AI, unblocking a $19.2M ARR opportunity for Google Cloud partners.
- Discovered and resolved a critical race condition in the Vertex AI backend (Java) that made requests to deploy the same model in the same Google Cloud project and region fail if they arrived within the same second; validated the fix with JUnit and Mockito test suites.
- Built the Model Garden Agent Assistant, a chat-based multi-agent system with Google's Agent Development Kit whose sub-agents autonomously discover open-source GenAI models, provision endpoints, deploy models, and run inference.
GoJavaPythonTerraformVertex AIADKTerraform Registry ↗Google Cloud Docs ↗
Google
Software Engineering Intern · Apache Beam
May 2024 → Aug 2024 · Seattle, WA
Built high-throughput data infrastructure for migrating legacy data into the cloud.
- Built a schema-aware Apache Beam transform in Java that can process 3 billion+ CSV records/hour on Google Cloud Dataflow.
- Built in safeguards for data integrity and fault tolerance by validating schema constraints ahead of execution and routing malformed records to a Dead Letter Queue.
JavaApache BeamCloud DataflowTerraform
StyleSyncs
Co-Founder & AI Engineer
Jan 2026 → Present · Waterville, ME
Co-founded an AI-powered virtual try-on platform with 100+ registered users and $9K+ raised through pitch competitions.
- Built the platform's AI features, including an image generation pipeline on Gemini 3 Pro Image for static virtual try-on and digital avatars.
- Developing the platform's flagship real-time video try-on feature, which overlays clothes on live camera feeds.
- Building a StyleSyncs iOS app in SwiftUI alongside the web platform.
ReactTypeScriptSwiftUIFastAPISupabaseGemini APICloud Runstylesyncs.com
Colby College
Undergraduate Teaching Assistant · CS
Sep 2024 → Present · Waterville, ME
Selected by faculty to mentor students in 10+ CS courses, including intro programming, Neural Networks, and Computer Organization.
- Support a cohort of 100+ students in courses like Data Structures & Algorithms, running code reviews and debugging sessions in Python, Java, C, and VHDL.
PythonJavaCVHDLNeural Networks

02 · ls ~/work

Selected work

Four projects I've built. Each one links to a case study with more detail.

main.tf

resource "google_vertex_ai_endpoint_with_model_garden_deployment" "gemma" {
  publisher_model_name = "publishers/google/models/gemma3"
  location             = "us-central1"

  model_config {
    accept_eula = true
  }
}

May 2025 → Aug 2025

Software Engineering Intern, Google

Vertex AI Terraform Resource

A Terraform resource for deploying AI models on Google Cloud

A Terraform resource I built during my 2025 Google internship that deploys generative AI models from Model Garden on Vertex AI (now Gemini Enterprise Agent Platform) with a single block of configuration. It's published in the official Terraform provider for Google Cloud, documented in the Google Cloud docs, and used by enterprises to deploy models today.

$19.2M ARR opportunity unblocked for Google Cloud partners who requested it

Official published in the Terraform provider for Google Cloud

read case study →

Jan 2026 → Present

Co-Founder & AI Engineer

StyleSyncs

AI-powered virtual try-on for fashion

A virtual try-on platform I co-founded that shows shoppers how clothes look on them. I built its AI features, including static try-on and digital avatars, and I'm currently working on real-time video try-on. It has 100+ registered users and has raised $9K+ through pitch competitions.

100+ registered users

$9K+ raised in pitch competitions

read case study →

Oct 2025 → Present

Lead Engineer

Mule Mart

An online marketplace for the Colby College community

A full-stack web app where verified Colby College students can buy, sell, and donate items. I started it with three other students, and as lead engineer I wrote most of the backend, including its semantic search feature and image storage service built on AWS S3.

Semantic search ranks listings by meaning instead of shared keywords

S3 private image storage, with uploads and downloads through presigned URLs

read case study →

Aug 2026 → Present

Tech Lead

Sankofa Lab website

Website for a STEM education non-profit in Africa

The website for Sankofa Lab, an educational non-profit that makes STEM learning accessible to pre-tertiary students in Africa. It's a fast static site built with Astro and hosted on Cloudflare Pages. A Cloudflare Worker in TypeScript handles the contact form and a question box where students can send questions to the team.

Static Astro site on Cloudflare Pages for quick load times

CI/CD GitHub Actions deploys the site and the Worker on every push

read case study →

also shipped

Vision Assistant

Offline-capable OCR device on Raspberry Pi for visually impaired users. Hybrid Cloud Vision / Tesseract pipeline with text-to-speech and ultrasonic proximity feedback.

Raspberry PiComputer VisionGCP

Model Garden Agent Assistant

A chat-based multi-agent system I built at Google with the Agent Development Kit. From a natural-language prompt, its sub-agents find open-source AI models in Model Garden or on Hugging Face, deploy them to endpoints, and run inference.

PythonGoogle ADKAI Agents

03 · tail -f roadmap.log

What I'm working on

A few things I'm actively building to strengthen my expertise in distributed systems, mobile development and harness engineering.

Mule Mart 2.0: Go microservices

in progress

I'm rebuilding Mule Mart's backend as event-driven Go microservices ahead of a relaunch, with separate services for listings and orders, messaging, and search. A GraphQL gateway gives the new React frontend a single API, and it talks to the services over gRPC. The messaging service uses WebSockets, so buyers and sellers can chat in real time. Semantic search runs in a FastAPI service that stores listing embeddings in PostgreSQL with pgvector. When something happens, like a new order, services notify each other through Google Cloud Pub/Sub. Everything runs on Cloud Run.

gRPC

GraphQL

React

FastAPI

PostgreSQLpgvector

ONNX

Pub/Sub

Cloud Run

OpenTelemetry

Pocket agent: on-device AI for iOS

up next

A native iOS app in Swift that will run open-weight models like Gemma 4 E4B fully on-device with MLX. The model will use tool-calling to work with Calendar, Reminders, and Shortcuts. Background triggers will let it check on things and notify you proactively. Because inference runs on-device, it needs no server or API key and works offline.

Swift

SwiftUImlx-swiftEventKit

App IntentsBackgroundTasks

Agent harness in Rust

queued

An agentic coding harness in Rust that works with any open-weight model. It will have its own agent loop, tool runtime, context management, and sandboxed execution, with a terminal UI built on ratatui. I plan to test it with my own Python eval suite, then benchmark it on a subset of SWE-bench Verified.

Rusttokioratatui

Docker

Python

04 · skills --list

Skills

Languages, frameworks, and infrastructure I've used in production.

~/francis-ohara/skills

❯ skills --group languages

# languages I write code in

Python

Java

Swift

TypeScript

JavaScript

C/C++

SQL

05 · git log --author=francis

GitHub

Recent activity from my GitHub, updated hourly.

2,704 contributions in the last year

refreshed hourly

NovDecJanFebMarAprMayJunJulAugSepOctNovMonWedFri

lessmore

2,704

contributions

past year

current streak

Nov 3 – Nov 6

longest streak

Aug 12 – Sep 6

public repos

followers

215

days active in the last year

2019

on GitHub since

coding-practice

This repository is a living collection of my journey through software engineering, featuring everything from competitive programming solutions to modern web application experiments.

Jupyter Notebook 1 0

vision-assistant

Low-cost prototype of an assistive device for reading text from a user's surroundings.

Jupyter Notebook 1 0

CS232-computer-architecture

Code I wrote for the Colby College computer science course CS232: Computer Organization.

VHDL 1 0

francis-ohara

My GitHub profile README

JavaScript 0 0

view full profile on GitHub

06 · ./connect.sh