This is my professional portfolio showcasing my academic achievements, certificates, and projects in electrical engineering, physics and computer science.
P. A. Sayegh, H. Déo, Y. Abou Rabii, and A. Couairon. European Journal of Physics (2026).
We built a scaled-down maglev train from permanent-magnet strips and measured how its levitation height changes with load. We derived a closed-form model of the magnetic force from first principles that matches 24 measurements with a single fitted parameter (R2 = 0.96), explained its vertical stability through Earnshaw's theorem, and scaled the model up to the full-size Transrapid TR08.
Read the PaperSelf-initiated work outside of coursework, in analog IC design, power electronics, and embedded hardware.
Team tapeout project in the IEEE SSCS Chipathon, designing a PPG analog front-end for heart-rate sensing on the open-source GlobalFoundries GF180MCU PDK. The front-end chains a pulsed LED driver, a capacitive-feedback transimpedance amplifier built around a folded-cascode OTA, and active filtering with a fully differential amplifier. My blocks include a pulsed LED driver (triode NMOS current-set device plus a switch device, 8 mA peak) and a 25 µA / 5 µA NMOS bias current mirror laid out in a common-centroid, guard-ringed array (83.4 × 9.0 µm). Both layouts are DRC clean and LVS matched, and schematics and layouts for the full design are complete.
View Team RepositoryDesign of a 200 kHz buck converter from specification to component selection: inductor and output capacitor sizing from ripple targets, conduction-loss and efficiency estimates, and continuous- versus discontinuous-conduction checks across load. The design is being verified against both an LTspice model and an independent cycle-by-cycle Python model, with a hardware build and closed-loop control planned next.
My first step from theory into hardware, built to learn how integrated circuits and embedded software come together on a real board. I paired an Adafruit Feather M0 (ATSAMD21 ARM Cortex-M0+ microcontroller) with an Ultimate GPS FeatherWing, which carries a MediaTek GPS receiver IC, powered from USB or a 3.7 V, 3000 mAh LiPo cell. In Arduino C++ I read the receiver's NMEA sentences over a serial (UART) link, parsed them with the Adafruit GPS library, and printed latitude, longitude, altitude and satellite count to the Serial Monitor once a valid fix was acquired. On an outdoor test the board reported a fix at 48°42.74′ N, 2°13.01′ E (171 m altitude, 3 satellites), matching my location at École Polytechnique in Palaiseau. I originally conceived it as a smart dog collar; adding a wireless link to a phone was left for future work. Along the way I learned to read IC datasheets, work with serial protocols and 3.3 V logic, and debug hardware outdoors, which is what pushed me toward circuit design.
View on GitHubHigh-frequency magnetics course project to design, build and characterize a resonant inductive wireless power transfer (WPT) system that lights an LED circuit on a toy car as it rolls over a transmitter coil. The preliminary targets are 100 kHz operation, 20 W delivered (12 V square-wave drive, 5 V / 4 A rectified output) and at least 70% efficiency. I am sizing the magnetics analytically (turns, wire gauge, gap, core and copper loss) for two core families, a TDK P22/13 pot core and toroidal cores, and comparing how core geometry affects coupling, losses and efficiency. Next steps are Maxwell finite-element simulation, winding the coils, tuning the series resonant capacitors and the rectifier, and measuring delivered power and efficiency versus gap and misalignment against a coupled-coil circuit model.
Replication of a published PID-fuzzy speed controller for a three-phase brushless DC motor (Energies 2024, 17(6), 1311), which uses sinusoidal commutation with third-harmonic injection and estimates speed from the timing of Hall-sensor edges. I am building the Simulink model of the drive, with a hardware implementation planned on an STM32 NUCLEO-G474RE and an X-NUCLEO-IHM07M1 three-phase driver board (L6230) running a Hall-sensor BLDC motor with an inertia disc.
An embedded AI system that performs real-time American Sign Language fingerspelling recognition entirely on-device, with no cloud dependency. Running on a Seeed Studio XIAO ESP32S3 Sense microcontroller, SignSpeak uses a 16,349-parameter depthwise-separable CNN (67.4 KB, INT8 quantized) achieving 97.4% accuracy on the Kaggle ASL benchmark, with end-to-end inference latency under 110 ms. The system outputs recognized letters to an OLED display with a 3-tier LED confidence indicator.
View Report Watch Demo View on GitHubDesign of a high-speed digital-to-analog converter targeting wireline transmitter applications, delivering ≥1.5 Vpp,diff into a 100 Ω differential load with SFDR exceeding 66 dB at the TT corner. The architecture uses a 3+5 segmented current-steering topology with thermometer-encoded MSBs, per-cell 6-bit trim DACs for ±10% mismatch correction, and retiming latches reducing inter-switch skew to ~1–2 ps. DNL remains within ±2×10−4 LSB across all three process corners.
View ReportDesign of a fully differential switched-capacitor sample-and-hold amplifier with 2× gain, intended as the residue amplification stage in a 7-bit 16 MHz pipelined ADC. The core is a folded-cascode OTA achieving 56.64 dB DC gain, 71° phase margin, 33.1 dB gain margin, and 167 MHz closed-loop bandwidth, with a settling error of 0.09% and total power consumption of 486.72 µW. A switched-capacitor CMFB circuit stabilizes the output common-mode at 0.9 V, achieving 333.4 dB CMRR.
View ReportDesign of a two-stage GaAs FET LNA on Duroid 5880 substrate targeting the Starlink customer terminal downlink band. The NE3210S01 HEMT was selected for its 0.34 dB NFmin at 12 GHz. The first stage is noise-matched to minimize Friis cascade noise figure; the second stage is gain-matched using Smith chart techniques. The final design achieves S21 > 23 dB, NF < 0.9 dB, S11 < −12 dB, and S22 < −14 dB across the band, with unconditional stability (µ > 1) from 0.1–20 GHz, IIP3 = 1 dBm, and 100 mW DC power.
View ReportThis report presents a digital signal processing (DSP) approach to runway detection for autonomous aircraft landing systems. The project leverages fundamental DSP techniques including edge detection, Hough transform, and frequency domain analysis to identify and track runway structures from aircraft camera imagery. The proposed system processes visual data through a multi-stage pipeline designed to extract geometric and textural features characteristic of runway environments.
View Report View on GitHubDesign of a two-stage Miller-compensated operational transconductance amplifier in 0.25 µm CMOS achieving 95 dB open-loop gain, 81.5° phase margin, 407 kHz closed-loop bandwidth with 10 V/V gain, and 13.7 µV output noise while consuming 247 µA from ±1.25 V supplies.
View ReportA physics-based approach for extracting device parameters and developing an all-region DC model for MOS transistors with uniform substrate. A systematic parameter extraction methodology is employed, utilizing transconductance-based threshold voltage extraction, geometry correction via maximum slope extrapolation, and mobility degradation characterization. The model incorporates effective mobility dependence, body effect, and accounts for short-geometry effects.
View ReportThis paper explores the properties of nonlinear acoustic strain wave propagation in solid crystals, integrating both experimental data analysis and numerical modeling. By introducing the Korteweg- De Vries (KdV) equation, we describe soliton propagation within certain nonlinear systems. The study simulates several physical scenarios, including the compression of a sequence of acoustic pulses, reflection at a free interface, and acoustic attenuation. These simulations provide insights into the complex behavior of nonlinear acoustic waves, which could have significant implications for the design and analysis of advanced materials and acoustic devices. This project was conducted at Laboratoire des Solides Irradiés (LSI) under the supervision of Dr. Vasily Temnov.
View ReportRogue waves are rare, unpredictable surface waves that exceed twice the height of surrounding waves and can appear suddenly from unexpected directions. Unlike tsunamis, these localized phenomena—sometimes surpassing 30 meters—arise from nonlinear energy focusing and pose significant risks to even the largest ships. This study focuses on training PINNs to solve the self-focusing Non-Linear Schrödinger equation and model the balance between nonlinearity and viscosity. This internship took place at LadHyX under the supervision of Prof. Christophe Josserand in January-March 2025.
View ReportThis project implemented several machine learning concepts to predict travel insurance purchases, focusing on key features like Annual Income and Travel History. Models like Logistic Regression, KNN, Neural Networks, and Decision Trees were trained on a dataset of 2,000 samples. Despite class imbalance, Decision Trees and Gradient Boosting provided the best results with an accuracy of 83.4%. The model helps insurance companies target potential buyers efficiently, improving marketing and resource allocation.
View Presentation View in GitHubWe developed a WebApp using Object-Oriented Programming in C++ and UI design in QML. Concept is a note-taking application that allows you to take notes in Markdown format and organize them in a file list structure. The application is written in C++ and uses the Qt framework for the GUI. This project is a CMake project and it is available for Windows, Linux and MacOS.
View in GitHubLaboratory project (LAB103) carried out with Vlad-Ștefănuț Radu in September 2022. We built two coin-cell lithium-ion batteries (CR2025 and CR2032 formats) from scratch: graphite anode and LiMn2O4 cathode pastes bound with PTFE on stainless-steel mesh current collectors, a lithium triflate / propylene carbonate electrolyte, and a microfiber separator, sealed with a hydraulic press. After an unmeasured first cycle, we ran four charge–discharge cycles per cell, recording current, voltage and time to compute capacity and Coulombic efficiency (mean 70.6% and 58.5% for the two cells). We propagated measurement uncertainties to a 99% confidence level, compared our results with published values for commercial cells, and analyzed the likely causes of the gap, such as residual moisture and variability in electrode preparation.
View ReportThis study aims to analyze the impact of social media on the outcome of the last parliamentary elections in Lebanon which took place in May 2022.
View Essay
Email: pas2232@columbia.edu
Phone: +16466790959
GitHub: https://github.com/peter-sayegh
LinkedIn: https://linkedin.com/in/peterasayegh