Portfolio — 2026

Lev Sosani

AI tools for BP's trading desks. Founding engineer at ExitPath, an operating system for M&A consulting firms.

I studied computer science and economics at UIUC, and now I'm in the Analytics Development Program at BP Trading in Chicago. Most days I'm figuring out where AI can actually help the traders around me, and then building it. On the side, I still help out at ExitPath and work on projects like SweepDreams, a free tool that helps fellow Chicagoans avoid street-sweeping tickets.

View selected work GitHub LinkedIn Email

Based in: Chicago, IL
Currently: BP Trading & Shipping
Status: Open to select projects

01 — Press

My open-source street sweeping alert system for Chicago, as featured on WGN-TV.

WGN-TV · August 2026

Chicago issues over 180,000 street sweeping tickets a year at $60 each, and the city doesn't warn anyone. After finding a wet $60 ticket waiting on my windshield, I wanted to help make sure that didn't happen to anyone else. You enter your address and it notifies you before your block is scheduled. It's free and open source, and since launching in May 2026 it has reached 1,000+ users and 45,000+ organic impressions.

The backend is Python and Flask on Postgres. It pulls ward section boundaries from the City of Chicago's open data portal, spatially joins each subscriber's address to its sweeping schedule, and sends the alerts on a schedule. Fully open source with no ads and no tracking; just an app to make Chicago a little better.

sweepdreams.org →Code Watch on YouTube ↗

02 — About

Curiosity and a love of building with AI, pointed at a trading floor.

Professional

I'm in BP's Analytics Development Program in Chicago, building tools that help our traders and analysts make the right call when it matters most. I help lead weekly S&D balance meetings, and I spend the rest of my time on one question: where can AI and software do the most for the people sitting around me?

So far the answers have been LLM-based systems that automate high-volume reporting for teams in Chicago, Houston, London, and Singapore, saving over 300 hours a year; a RAG system over 16 years of internal refinery data that answers questions about 3x faster than the enterprise tools it replaced; and a rebuilt DOE release pipeline that reaches 1,500+ people in under five seconds. I presented that work 1:1 to BP's CTO, which helped boost funding for AI across our desks.

Before BP, I was a founding engineer at ExitPath, an operating system for M&A consulting firms. I joined as one of the first engineers, shipped v1 in under 16 weeks, and helped grow it to 10+ enterprise clients and around $100K ARR before it was acquired in 2026. I still help out on weekends where I can.

Personal

I graduated from the University of Illinois Urbana-Champaign in May 2025 with a B.S. in Computer Science and Economics and a 3.98 GPA.

Most of my side projects start with something that I wish I had in my daily life. SweepDreams came out of Chicago's street sweeping tickets. HomeBase came out of needing four different apps to control one apartment. TradeScrape came out of wanting to test whether the trade ideas people pay for actually work.

Outside of work I love spending time with my family, reading books like 'Build' by Tony Fadell, and trying out whatever new tools or LLM models came out that week. The rest of my time goes to planning my travels and hitting the gym.

Lev learning about CCC's AI insurance software during the 2025 City Scholars Co-Op — Learning about CCC's AI insurance software during my 2025 City Scholars Co-Op

Education

Lev throwing his graduation cap at the University of Illinois Urbana-Champaign — Graduation day at UIUC, May 2025

Class of 2025

B.S. in Computer Science and Economics

University of Illinois at Urbana-Champaign

GPA 3.98/4.00

LAS James Scholar
Dean's List 2021–2024

Experience

June 2025 – Present

Chicago, IL

Analytics Development Program

BP Trading & Shipping

I'm an analyst on BP's trading floor, and most of my impact has come from building AI tooling for the analytics teams around me. I've designed and deployed LLM-based systems that automate high-volume reporting workflows, saving teams in Chicago, Houston, London, and Singapore over 300 hours a year, and built a RAG system on Databricks that lets analysts query 16 years of internal refinery data about 3x faster than the previous enterprise tools. I also automated distribution of the DOE weekly data release, taking release-to-send time from 3 minutes to under 5 seconds for 1,500+ recipients, including 100+ traders and analysts. I presented this work 1:1 to BP's CTO, which secured expanded funding for AI across the trading analytics organization, and I lead the weekly S&D balance meeting for the distillates, gasoline, and crude desks.

September 2024 – Present

San Francisco, CA (Remote)

Founding Engineer

ExitPath

One of the first engineers at ExitPath, an operating system for M&A consulting firms. I architected the core platform, shipped v1 in under 16 weeks, and helped scale it to 10+ enterprise clients and about $100K ARR before its acquisition in 2026. I built the MCP server layer that let AI agents operate the platform, wrote an automated triage service with Claude Code that classified and routed inbound bug reports and feature requests to the backend team, and led a zero-downtime migration from Firestore to Supabase that cut infrastructure cost by 60% and average query latency by 25%.

August 2024 – December 2024

Chicago, IL

Fundamentals Modeling & Innovation Intern

BP Trading & Shipping

My first time on the trading floor. I built Python models on economic data for BP's desks, moved them onto faster frameworks in Python and Dataiku, and added OLS regressions to predictive models used in risk management. I went back to school in December and returned full-time the following June.

May 2023 – March 2024

Manhasset, NY

Full-stack Software Engineer

DuelDog

I led a cross-functional team re-architecting a legacy .NET platform into a cloud-native system on AWS. We broke a monolithic backend into independently deployed microservices behind REST APIs, which cut page load times by 30% and improved real-time data sync across the product.

May 2022 – August 2022

Champaign, IL

CS 128 Course Assistant

UIUC Computer Science

I helped teach C++ to 115+ students in UIUC's intro CS sequence, running code reviews and one-on-one debugging sessions across 200+ student projects.

03 — Selected Work

Some of my favorite side projects.

SweepDreams

Civic Tech

A free, open-source service that alerts Chicago residents before street sweeping so they don't get ticketed. The city writes 180,000+ sweeping tickets a year at $60 each. Since launching in May 2026 it has reached 1,000+ users and 45,000+ organic impressions, and WGN-TV ran it as a lead story in August 2026. The backend ingests City of Chicago ward section boundaries, spatially joins them to each user's address, and sends scheduled notifications at close to zero running cost.

Python
Flask
PostgreSQL
Vercel
GeoJSON

Code Live WGN-TV segment ↗

Finn AI

AI · HackIllinois '25

A personal-finance chatbot built at HackIllinois 2025, where it took 2nd place in the CAT Digital track. It links a user's bank account through Plaid and answers questions about their spending in plain English. LangChain and GPT-4o mini handle the conversation, with a Flask backend, a React and TypeScript frontend, and MongoDB.

React
TypeScript
Flask
LangChain
GPT-4o
Plaid
MongoDB

Code Demo ↗

Self-Driving Car Simulator

Computer Vision

A small self-driving car built on a Raspberry Pi and Pi Camera. OpenCV detects lane lines in each frame and a steering loop keeps the car centered on a marked course, with no other sensors.

Raspberry Pi
Python
OpenCV
Computer Vision
Pi Camera

Code

TradeScrape

Trading & Finance

A self-funded algorithmic trading agent that ran up to 100 trades a day through the Tradier brokerage API on AWS EC2, with real-time market data ingestion and automatic recovery from failures. A companion pipeline scraped 500+ trade ideas a week from paid trading communities, classified them with GPT-4, and backtested them against Polygon.io options data, which improved trade selection accuracy by 45%.

Python
React
SQL
Redis
AWS
GPT-4

Private

HomeBase

IoT & Automation

A home automation hub on a Raspberry Pi 4 that controls the Tesla API, smart lights, weather data, and a live stock portfolio from one interface. Each integration is its own Flask microservice, so adding a device doesn't touch the others.

Linux
Flask
Python
TypeScript
Docker
Raspberry Pi

Code

Handtracking-Colormatch

Computer Vision

A real-time hand tracking app. MediaPipe extracts hand landmarks from a webcam feed and OpenCV maps hand position and gestures to color selection.

Python
OpenCV
MediaPipe
Computer Vision

Code

Govee-Python-Library

IoT

A Python library for controlling Govee smart lights and plugs. It wraps Govee's REST API in simple methods so devices can be scripted in a few lines instead of through the phone app.

Python
API Integration
IoT
Home Automation

Code

Wikipedia Connectivity Visualizer

Data Analysis

A C++ tool that analyzes the Wikipedia link graph using Stanford's SNAP datasets. It finds shortest paths and clusters between articles and renders the results as an interactive visualization.

C++
Data Visualization
Stanford Datasets
Network Analysis

Code

Get in touch.

Email: sosanilev10@gmail.com
Location: Chicago, IL

Available for

Applied AI, software engineering, and fintech opportunities