• Embedded products: ESP32-S3 firmware, PlatformIO, LVGL, SPI displays, rotary input, WiFi provisioning, OTA, and persistent device state.
• Robotics and autonomous systems: ROS 2, vehicle dynamics, controls, sensors, CAD, 3D printing, and lab-scale hardware integration.
• AI + backend systems: Python services, FastAPI, SQLAlchemy, Redis, tests, model routing, Telegram workflows, and provider abstractions.
• Safety-aware tools: deterministic checks around AI output, explicit failure modes, risk gates, cache fallbacks, and human confirmation.

TokenJar is a small desk device that shows Anthropic and OpenAI usage without opening a browser tab. It combines a 2-inch ST7789 LCD, an ESP32-S3 SuperMini, an EC11 rotary encoder, a 3D-printable enclosure, and firmware that talks to multiple usage sources.

What it shows technically:

• Real embedded product loop: hardware, firmware, UI, enclosure, setup flow, and troubleshooting are all documented.
• Split runtime architecture: main loop drives LVGL and encoder interactions while a FreeRTOS API task refreshes provider data.
• Provider integration: Anthropic Admin API, OpenAI Admin API, Claude.ai session usage, and ChatGPT/Codex usage paths can be configured independently.
• Device reliability: NVS-persisted credentials, cached snapshots, WiFi reconnect handling, mDNS as tokenjar.local, OTA updates, idle dimming, and runtime settings portal.

C++ PlatformIO LVGL TFT_eSPI ArduinoJson ESP32-S3 FreeRTOS-style tasking

SignalForge AI is a Telegram-first market research assistant. The important design choice is separation of responsibility: Python calculates indicators, portfolio state, paper-trade sizing, and deterministic risk; AI explains results and routes intent, but does not override the risk manager or execute real orders.

What it shows technically:

• Layered backend: Telegram and FastAPI entry points, service layer, quant modules, risk modules, AI modules, SQLAlchemy persistence, and Redis cache support.
• Deterministic risk controls: stop-loss, thesis, allocation, account-risk, single-stock exposure, margin, shorting, options, and penny-stock rules reject unsafe paper trades before persistence.
• Provider abstractions: yfinance demo provider today, with Alpaca/Polygon/Finnhub/Tiingo-style production provider boundaries already reflected in the architecture.
• Testable workflow: tests cover Telegram parsing/handlers, indicators, scanner behavior, backtesting, risk sizing, portfolio services, allocation, and AI model routing.

Python FastAPI SQLAlchemy Redis APScheduler pandas yfinance OpenAI Telegram Bot API pytest

An earlier multi-agent research framework with 11 LLM agents, a 6-phase debate pipeline, local technical indicators, risk debate, chart generation, and Telegram delivery.

Python OpenAI/Anthropic yfinance Telegram

Photo geolocation pipeline using EXIF extraction, visual reasoning, hypothesis generation, external verification, final scoring, CLI usage, and Telegram bot workflow.

Python Vision AI Reasoning Telegram

Personal portfolio with robotics, vehicle controls, mechanical design, autonomous systems, capstone work, maker projects, and project photography.

TypeScript Next.js Tailwind Robotics

Mechanical design -> sensors -> firmware -> data pipeline -> AI reasoning -> usable product

I am comfortable moving across that path: designing parts in CAD, wiring sensors, writing embedded UI code, building Python services, and shaping the final user experience so the system is not just clever, but usable.

• Incoming M.S. Robotics student at the University of Michigan.
• Completing B.S. Mechanical Engineering at RPI, GPA 3.87 / 4.0.
• Undergraduate researcher at XAL Research Lab, working on autonomous-vehicle systems for a Can-Am X3 platform.
• Tesla engineering intern in 2025, focused on CAD redesign, thermal/manufacturing assessment, and autonomous-driving sensor mounting.
• Seeking Summer 2026 robotics, autonomous systems, controls, embedded, hardware, or AI engineering internships.