MOSAICLEDGER
TartanHacks team of 3: Plaid + Capital One Nessie finance dashboard with D3 treemap spending viz, subscription detection, hosted MCP tools for AI coaching, XRPL round-up savings, and Vitest fuzz + Playwright E2E CI.
_ ____ ____ ___ _ _ _ __ __ _ _ _ _ ___ _____
/ \ | _ \| _ \|_ _| / \ | \ | | | \/ | | | | \ | |/ _ \__ /
/ _ \ | | | | |_) || | / _ \ | \| | | |\/| | | | | \| | | | |/ /
/ ___ \| |_| | _ < | | / ___ \| |\ | | | | | |_| | |\ | |_| / /_
/_/ \_\____/|_| \_\___/_/ \_\_| \_| |_| |_|\___/|_| \_|\___/____|
Carnegie Mellon BS+MS Electrical & Computer Engineering student with a Robotics minor, building production software across React, Spring Boot, AWS, systems programming, networking, AI tooling, and security.
Note: All activity data is using UTC time.
TartanHacks team of 3: Plaid + Capital One Nessie finance dashboard with D3 treemap spending viz, subscription detection, hosted MCP tools for AI coaching, XRPL round-up savings, and Vitest fuzz + Playwright E2E CI.
100% deepfake traceability via LSB watermarking (<0.4% pixel change) and RSA-2048/4096 signatures, with automated PKI and Docker multi-stage builds fixing cross-platform deployment across a 5-framework ML stack.
MV3 extension detecting and optionally blocking 1×1 pixels, sendBeacon, and scripted network calls via fetch/XHR/Beacon hooks, heuristic Low/Medium/High risk scoring, and a React UI with Observe/Soft-Block/Strict modes.
HTTP/1.1 server built from scratch with low-level Python sockets, full method/status support, byte-range video streaming, and 200+ concurrent clients.
Full-stack chat app with authentication, friend lists, REST APIs, responsive UI, websockets, and real-time group chat for 20+ users.
Thread-safe disk buffer pool with page fetch, eviction, pin-count tracking, LRU-K replacement, background disk scheduler, promise I/O callbacks, and Read/WritePageGuard RAII for concurrency-safe access.
Custom 50-class US landmark dataset comparing ResNet-50 vs ViT-B16 with five loss functions, MixUp, and OOD detection. ViT-B16 + Class-Balanced Loss + MixUp reached 75.72% Top-1 (+10.57pp Tail-F1 vs baseline).
Real-time Canny pipeline on GPU with fused kernels that cut intermediate global memory round-trips and launch overhead. Owned image frontend, preprocessing, CUDA stream management, and kernel launches.
ECE senior capstone: low-cost semi-autonomous multi-robot swarm for urban disaster survivor search with LiDAR, UWB-IR ranging, IMU fusion, and frontier-based exploration. Owned collaboration logic, exploration, obstacle avoidance, and the 2D operator dashboard.
Fuses 6-DoF AprilTag pose with 3-DoF IMU orientation to drive a live ARENA scene; Pi Camera OpenCV checkerboard calibration, tag36h11 detection, and pose (R, t) estimation from images.
LIDAR + odometry occupancy grids for >90% arena navigation accuracy; learning-based object detection, A*/RRT* motion planning, and behavior trees for task-level execution.
Custom Fusion 360 PCB with I2C, LCD, UART, and servo drivers for user-mode programs; PID motor control with speed regulation and UART telemetry to a Python visualization script.
I'm Adrian Muñoz, a Carnegie Mellon electrical and computer engineer who likes building across the stack: backend services, interfaces, infrastructure, embedded/systems code, and robotics-adjacent software.
My recent work includes shipping a production admin UI at Atlassian, building AWS data-processing infrastructure at Amazon, evaluating AI coding models at Outlier, and building projects that span networking, cryptography, browser privacy, chat systems, and database internals.
// personal photo archive loaded from committed browser-safe images.