Software engineer · Distributed systems · Cloud infrastructure

Hi, I’m Leonardo Sola!

I build reliable software for data-intensive and physical systems—from streaming vector search to rocket avionics and computational fabrication.

Incoming SDE @ AWS Amazon FSx
View featured work View résumé
GitHub LinkedIn
Portrait of Leonardo Sola

M.S. in Computer ScienceBoston University

Projects

Featured work

Three vector-state architectures evaluated
Remote HoloStream Milvus
Embedded DuckDB shards Top-K reducer
Peer-to-peer DuckDB worker Peer workers
CorrectnessLatencyThroughputScaling

Distributed systems · Vector search

HoloStream Vector Search

On a five-person team, we added real-time KNN search to a Go streaming engine and benchmarked three vector-state architectures on SIFT1M across one, two, and four workers.

  • Go
  • gRPC
  • DuckDB
  • Milvus
View source

Selected work

Other projects

Three-dimensional tumor models and source MRI scans

Computational fabrication · Medical imaging

MRI-Based 3D Printed Tumor Models

Our team built a pipeline that transforms segmented MRI scans into printable 3D tumor models, giving doctors a physical way to inspect complex anatomy before an operation.

  • Python
  • Marching Cubes
  • 3D fabrication
View source
BURPG rocket avionics control panel

Embedded systems · Aerospace

BURPG IREC Rocket

I helped build and test a Teensy 4.1-based flight computer that controlled telemetry, ejection charges, data recording, and parachute deployment.

  • Teensy 4.1
  • Telemetry
  • Hardware testing
Natura scenic route planner map and interface

Geospatial software · Computer vision

Natura

A scenic route planner that combines open map data and street-level imagery to score and compare the most scenic drives between two points.

  • OpenStreetMap
  • Mapillary
  • OSRM
View source
EasyNote note-taking project logo

Hackathon winner · Education software

EasyNote

I led a three-person team building a Chrome extension that turns YouTube videos into study guides. EasyNote won its division at BostonHacks.

  • Python
  • Node.js
  • Chrome extension
View Devpost

Additional Projects

Datathon project

Predictive Model

Worked with a team to prepare, visualize, model, and evaluate a real-world dataset using pandas and scikit-learn.

View source

AI web application

ChatGPTInsta

Built a React interface and Python backend that generate social-media captions, hashtags, and post descriptions from user prompts.

Experience & education

Next · Starting 2026

Incoming Software Development Engineer

Amazon Web Services · Amazon FSx

I’ll be joining the Amazon FSx team as a Software Development Engineer, building cloud storage technology at AWS.

2023–2024 Robotics

Robotics Instructor

Robotfun Academy

Taught engineering and software labs for students in grades 4–8 using Scratch, Python, REV, and LEGO robotics. I helped students turn basic physics and design principles into working robots.

2023 Education

Mathematics Instructor

Rocket Club Math

Collaborated on K–8 lesson plans with different levels of difficulty, tailored material to individual grades, and helped streamline the team’s file storage and organization.

Capabilities

Tools I’ve used to take projects from idea to working system.

Systems & infrastructure

Go, Java, Python, C/C++, Linux, Bash, gRPC, Docker, networking

Data & backend

DuckDB, Milvus, vector search, stream processing, pandas, scikit-learn

Other engineering

Embedded systems, robotics, React, JavaScript, HTML, CSS, Android Studio, SolidWorks, testing and debugging

Leonardo Sola outdoors

About me

I’m a software engineer who enjoys working on problems where performance, reliability, and thoughtful system design matter. I recently completed my M.S. in Computer Science at Boston University and will be joining AWS as a Software Development Engineer working on Amazon FSx.

My work has ranged from distributed data processing and vector search to embedded systems, robotics, and computational fabrication. My favorite languages to reach for are Go, Java, and Python, depending on the problem. Outside of engineering, I enjoy hiking, photography, and gaming.

Download résumé

Photography · 2019–2026

Creativity outside of the office.

Sunset over Tampa Bay Synchronized skater holding a solo pose Long-exposure light trail from an electric skateboard
View the full gallery

Get in touch

Want to connect?

I’m always happy to connect about software engineering, interesting projects, or anything else tech-related.

leonardozsola1@gmail.com
Email me Connect on LinkedIn
MRI slices compared with a reconstructed three-dimensional tumor Diagram of the MRI-to-3D-model processing pipeline

Computational fabrication · Medical imaging

MRI-Based 3D Printed Tumor Models

Our project explored a new way for doctors to visualize tumors before operating on a patient. We built a pipeline that takes pre-segmented MRI scans, extrapolates and orients the individual slices, and interpolates the spaces between them before using Marching Cubes to generate a printable 3D mesh. Although our work focused specifically on brain tumors, the same process could be applied to other collections of segmented MRI scans to create physical models that make complex anatomy easier to inspect and understand.

View source
BURPG IREC rocket avionics control panel

Embedded systems · Aerospace

BURPG IREC Rocket

I contributed to the design, production, testing, and debugging of a flight computer for the Boston University Rocket Propulsion Group’s IREC team. Built around a Teensy 4.1 microcontroller and a variety of Adafruit components, the system served as an integral part of the rocket’s avionics. It was responsible for coordinating critical flight functions, including recording and transmitting telemetry, controlling ejection charges, and supporting parachute deployment, all of which required the hardware and software to operate reliably under demanding flight conditions.

Distributed systems · Vector search

HoloStream Vector Search

On a five-person team, we integrated streaming KNN search directly into HoloStream and compared three ways to manage vector state: a remote Milvus service, embedded DuckDB shards with downstream MapReduce aggregation, and embedded DuckDB shards with peer-to-peer aggregation.

My contribution

I rewrote the backend RPC path so heterogeneous state backends could run across different workers. I also contributed the initial Milvus workflow, completion signaling, brute-force correctness tools, and experiment scripts used to collect latency and throughput measurements.

Testing setup

1M SIFT vectors
128 dimensions
10K query vectors
1, 2, 4 parallel workers

CloudLab experiments ran on c6620s machines with 28-core 2.1 GHz Intel CPUs, 128 GB of RAM, and 25/100 Gb network interfaces. Tests used K=5, input rates of 100 and 1,000 queries per second, exhaustive search, and HNSW.

Verified results

  • All three architectures produced correct top-K output against brute-force ground truth.
  • At four-worker parallelism and a 1,000-query input rate, exhaustive-search throughput was approximately 24 queries/sec for remote Milvus, 11 for MapReduce, and 9 for peer-to-peer aggregation.
  • With HNSW, MapReduce peaked near 287 queries/sec at two workers; peer-to-peer peaked near 277 queries/sec at one worker and declined as network fan-out increased.
  • Input rate changed which architecture had the lowest latency, while peer-to-peer had the highest overall CloudLab latency.
View source
Natura scenic route planner map and imagery interface

Geospatial software · Computer vision

Natura

Natura is a scenic route planner MVP that turns open map data and street-level imagery into the most scenic drive between two points. It retrieves road data from OpenStreetMap, samples a uniform grid of locations, matches those samples to Mapillary images, and extracts simple visual cues from each image to estimate the surrounding scenery. The pipeline then builds a heatmap, uses it to score route alternatives produced by OSRM, and presents the results on an interactive map where users can compare routes, inspect imagery coverage, and rerun scenic routing without downloading the same images again.

View source
EasyNote project logo

Hackathon winner · Education software

EasyNote

I led a three-person development team in creating EasyNote, a Chrome extension designed to help students study more efficiently. The extension takes a YouTube video, uses the AssemblyAI API to process its content, and generates a comprehensive study guide that can be delivered through the Google Docs API. We combined Python, Node.js, and HTML to build the project and connect its different services into a practical workflow. Our work earned EasyNote a division win at the BostonHacks Hackathon.

View Devpost
Datathon predictive-model data dashboard

Datathon project · Machine learning

Datathon Predictive Model

I worked with a team to analyze a real-world dataset and extract useful insights using pandas and scikit-learn. We applied the core stages of a machine-learning workflow, including data preparation, visualization, model implementation, and performance evaluation. Alongside the project, I attended data science and software engineering seminars that strengthened my understanding of these concepts and gave me an opportunity to apply them to a practical problem.

View source
ChatGPTInsta application interface

AI web application

ChatGPTInsta

ChatGPTInsta was created to help users streamline their social media presence with AI-generated content. I built an intuitive, visually engaging interface in React and connected it to ChatGPT through a Python backend, allowing users to submit prompts and receive generated captions, hashtags, and post descriptions. The project was also designed with future scheduling support in mind, giving users a path toward automatically publishing generated posts to Instagram instead of managing each post manually.