// SOFTWARE ENGINEER & WRITER
Building digital
experiences_
I'm Sonny, a Haitian American software engineer and student, driven by one question: how do I build things that actually matter?
That's led me to some interesting places. Starting in high school, I cofounded Recomputing and built Bright Eye, a mobile AI role play app with over 30,000 downloads, $15,500 in funding, and a lot of hard lessons about what it actually takes to ship something people use. Since then I've interned and built production systems at LinkedIn, Nvidia, Schneider Electric, Siemens, and most recently Lexi, an early stage startup where I was hired as the first intern to work on multilingual NLP for real-time clinical translation.
This site is split in two. Half of it is my engineering work: projects, things I've built, what I'm currently obsessing over. The other half is writing: on technology, society, and whatever I can't stop thinking about. Some of it is behind a code. You'll know why when you read it.
When I'm not building or writing, I'm hiking, traveling, chasing the next startup idea, or in the gym. Glad you're here.
// Experience
First intern hire — building multilingual NLP systems and the public mobile version of Lexi's medical AI translation service.
Key Achievements
- › Hired as the first-ever intern at Lexi, an early-stage medical AI startup
- › Developing multilingual NLP systems enabling real-time clinical translation
- › Building the public mobile version of the Lexi medical AI service
Built predictive analytics infrastructure and developer tooling for multi-cloud cost forecasting across AWS, Azure, and GCP.
Key Achievements
- › Built a Python-based predictive analytics service to forecast cloud usage/spend across AWS, Azure, and GCP
- › Prototyped a self-serve web portal (JavaScript) with API integrations to surface cloud-efficiency scores
Shipped a production RAG chatbot and an ML-powered confidentiality tagging system that cut manual review by 35%.
Key Achievements
- › Built and deployed a Python RAG chatbot (FastAPI) using the Glean API, Azure SQL, and OpenAI embeddings — reduced manual doc time by 10+ min per dataset
- › Built a scalable confidentiality tagging system using Logistic Regression (scikit-learn) + MiniLM, reducing manual review by 35%
Built AWS serverless fraud-detection tooling and led a Nuxt 2→3 migration saving months of future refactoring.
Key Achievements
- › Built an internal review system for fraudulent installer reviews using AWS Lambda (Python CDK), DynamoDB, S3, and a Superblocks frontend
- › Refactored 800+ lines from Nuxt 2 to Nuxt 3, saving months of future work and preserving business function continuity
Saved $225K+ in licensing costs with a Python desktop app and shipped a LLaMA RAG chatbot for internal documentation.
Key Achievements
- › Built a full-stack Python compiled executable for THB metric calculations (NumPy, SQLite, PyInstaller) for the Atellica DCA app
- › Reduced licensing costs by 60%, saving $225K+ compared to the previous solution
- › Developed a LLaMA-based AI chatbot with RAG through Elasticsearch, providing context-aware responses from internal documentation
Co-founded Recomputing and shipped Bright Eye — 30K+ App Store downloads, $15.5K raised, full stack from mobile to web.
Key Achievements
- › Grew Bright Eye to ~30,000 App Store downloads and $15,500+ in preseed funding
- › Engineered a bot creation backend using RAG, Supabase vector database, and OpenAI text embeddings
- › Built a Node.js LangChain/Express backend for secure API and credential management
- › Launched web client at brighteye.app with Next.js, serving 200+ desktop users
- › Implemented RevenueCat SDK subscription layer resulting in 12 paying subscribers
10+
Projects Shipped
1
Essays Written
4
Years Building
∞
Coffee Consumed
// Recent Work
Bright Eye
Platform for creating highly customizable AI chatbots for professional and entertainment use cases. Over 30K+ downloads and $15K raised in preseed funding.
BeatTheClock
Real-time personalized AI job-displacement risk platform that estimates how exposed your role is to automation and translates that into a simple, time-based warning system.
Fragmentation Algorithm
Graph-based gerrymandering detection tool that quantifies community fragmentation in congressional district maps, applied to Indiana's redistricting to compare enacted vs. proposed maps.