Ozan Ekame Pekgoz
Live: the sky over Boston right now

Ozan Ekame
Pekgoz

Computer engineering student at Boston University. I build hardware, write the code that runs on it, and get the right people in the room to finish it.

See my work Resume
GitHub LinkedIn

About

I was born in Istanbul, grew up in Barcelona, and moved to Boston for university. Ever since I was a kid, I've known that whatever I chose to do, I wanted it to make a real difference in people's lives.

When it came time to choose a degree, computer engineering felt like the right fit, not only because of how versatile it is, but because I knew it would challenge me and push me to grow further than I could imagine.

That's what I'm working on now. With my startup, Cohort Academy, I'm connecting students across campus online, and with FireDrone, I'm helping limit the damage from the wildfires that hit every year.

Languages

  • Spanishnative
  • Turkishnative
  • Englishnative
  • Catalanfluent
  • Frenchintermediate

Outside engineering

  • Piano
  • Guitar
  • Surfing
  • Basketball

Tools

Hardware
  • Verilog
  • Vivado
  • FPGA
  • Arduino
  • Embedded C
  • Cadence Virtuoso
  • I²C / SPI
  • Oscilloscope
  • Soldering
Software
  • C/C++
  • Python
  • JavaScript
  • HTML/CSS
  • WebAssembly
  • Java
  • MATLAB
  • Bash
  • Git
Ozan at a crosswalk in Akihabara, Tokyo, hands on his head and grinning

Education

Boston University

May 2027

B.S. Computer Engineering

Coursework
  • Computer architecture
  • VLSI design
  • Circuits
  • Electronics
  • Logic design
  • Software engineering
  • Algorithms
  • Probability

What I'm working on now

Two projects in progress this year.

Startup · since May 2026Live

Cohort Academy

A free online community for BU students in intro STEM courses. Each course gets a space with chat, exam prep guides and review sessions run by upperclassmen and TAs, with optional peer tutoring on top.

It started as a class project in spring 2026 that I co-founded with Oliver Zeman. I built the whole platform from scratch with AI coding tools, and I kept doing one-on-one interviews with students to find out what to fix and what to add next. We went from 0 to over 200 members in 3 weeks, and we're at 230+ now with about 30% of them using it every week.

  • On the web, on the App Store for iPhone, and as a Windows and Mac app
  • BU-only Google sign-in, live chat, events with automatic Google Meet links, and tutor booking
  • 83 exam prep guides and past syllabi for 85 courses
cohort.academy
Senior design · 2026-27Component testing

FireDrone

Wildfires are far easier to stop before they start. The goal is a system that uses satellite imagery and AI to flag areas at high fire risk, then sends an autonomous drone to check conditions on the ground, like how dry and hot it is.

It's my idea, and I lead our team of six: three mechanical engineers and three electrical and computer engineers. I set the plan and make sure everyone is getting their tasks done, and on the electrical side I'm working on the wiring and a custom sensor PCB, and reaching out to people who could use it in the field.

  • Now: component testing, checking each part on its own before it goes on the drone
  • Done: requirements and a warm-up mini-project as a team
MechanicalOliver Zeman, Shervin Boostani, Season Thapa
Electrical & computerAakash Kumar, Jeehan Zaman, Ozan Ekame Pekgoz

Projects

Class and team projects from the last two years. Click any of them for the details and who I built it with.

Experience

Jobs, internships and the things I do outside class.

May 2026 - present

Co-founder

Cohort Academy · Boston · with co-founder Oliver Zeman
  • Built cohort.academy from scratch with AI coding tools and shipped it on the web, the App Store and as a Windows and Mac app.
  • Run one-on-one interviews with students and turn their feedback into new features. 0 to 200+ members in 3 weeks, now 230+ with about 30% active each week.
  • Pitch Cohort to students and staff, and organize the upperclassmen who lead exam reviews.
Summer 2025 - present

Student Advisor, College of Engineering Volunteer

Boston University · EK100
  • Help incoming engineering students pick courses, understand major requirements and find campus resources.
  • Explain program structure to students and families from many backgrounds, sometimes in Spanish or Turkish.
May - Aug 2024

Sales and Product Management Intern

Dima Textile · Barcelona
  • Won a major new client by taking a suitcase of textile samples to their headquarters and presenting our product lines.
  • Managed their first orders through production and delivery, tracking them across teams in Excel.
  • Pitched prospective clients in Spanish, English and Turkish.
Jul 2022

IT Support Intern

Alliance Healthcare · Istanbul
  • Fixed hardware and software problems across departments and helped with server security maintenance and order-tracking systems.
Sep - Dec 2022

Assistant Coach, Men's Basketball

EISB
  • Ran practice drills and helped with game-day decisions.

Get in touch

I'm looking for internships in hardware, embedded systems or software. Email is the fastest way to reach me.

Email ozanekame@gmail.com
LinkedInin/ozanekame GitHubOzanEkame

Cohort Academy

Startup · Web, iPhone, Windows, Mac · May 2026 to now
Visit

BU sign-in Google accounts @bu.edu only Course spaces live chat past syllabi Prep guides 83 exam guides Live sessions Google Meet links RSVP + calendar Tutoring weekly hours first hour free Next.js · React · TypeScript · Supabase realtime · Vercel

The idea

Intro STEM courses at BU are big, and a lot of first-years get through them alone. Cohort gives every course a shared space where students can ask questions, find prep material and join reviews led by students who already took the class. Joining is free. Verified tutors can offer paid sessions, and each student's first tutoring hour is free.

How it started

It began as a final project for an innovation class in spring 2026, first as a Discord and Whop community. My co-founder Oliver Zeman and I kept it going over the summer, and the native platform went live at cohort.academy in September 2026, in time for the first exams of the fall. The iPhone app followed on the App Store at the end of October, along with a desktop app for Windows and Mac.

What I built

I built the whole thing from scratch with AI coding tools (Claude Code and Codex), and I designed and checked every piece of it myself. It has Google sign-in limited to BU accounts, realtime chat per course, a resource library, events with RSVPs, Google Meet rooms created automatically for each session, Google Calendar export, tutor booking with Stripe payments, push notifications, roles and an admin view. It runs on Next.js, React, TypeScript, Supabase and Vercel, the iPhone app is built with Capacitor and the Windows and Mac app with Electron.

The people part

Most of the work outside the code is talking to people. I've done over 20 one-on-one interviews and talked with more than 100 students to understand what they actually need, and that feedback decides what I build next. I gave a 2-minute pitch to about 300 engineering freshmen in an EK100 lecture and 76 of them signed up, so about 1 in 4. We went from 0 to over 200 members in 3 weeks, we're at 230+ now with about 30% active every week, and we won a $500 BU Wellbeing Innovation Seed Grant (top 15 of 43 proposals). Oliver drives a lot of the outreach too, from the campus flyer to the emails to incoming freshmen, and we bring on upperclassmen and TAs to run the reviews.

FireDrone

EC463/464 Senior Design · Team 06 · 2026-27
Satellite data imagery, weather fire history Risk model AI flags high-risk areas Dispatch autonomous drone launch On-site check dryness, heat camera survey Report risk alert return to launch the goal of the system · design in progress · under 400 ft · $500 to $750 budget

The goal

Most wildfire tools react after a fire is already burning. We want to catch the conditions before that: analyze satellite data on past fires and their causes, use it to flag high-risk areas, and send an autonomous drone to those areas to measure what's happening on the ground, such as dryness and heat. Which readings count as dangerous is part of what we still have to research.

Where we are

We're in the component testing phase, testing each part on its own before it goes on the drone. Before that we worked through the requirements, finished a warm-up mini-project as a team, and compared flight controllers (a Pixhawk 6C running ArduPilot, or a cheaper F405 board) for a drone with a small custom PCB to connect the sensors. The drone has to stay under 400 ft and fit a budget of roughly $500 to $750.

My part

FireDrone was my idea. I brought the six of us together and lead the team, making sure we're all getting our tasks done. I'm on the electrical side, handling most of the wiring and the sensor PCB, and I'm looking for a first client who flies drones over fire-prone land. For the mini-project I wired and soldered the motor driver to the microcontroller and drew the wiring schematic.

Team: Oliver Zeman, Shervin Boostani and Season Thapa (mechanical); Aakash Kumar, Jeehan Zaman and Ozan Ekame Pekgoz (electrical and computer).

Color sorting machine

EK210 · Arduino · Jan-May 2025 · Class award
Code
Sorted Skittles in colored bins under the machine

The brief

Our EK210 team had one semester to build a machine that sorts objects by color with as little human help as possible, using cheap parts. We picked Skittles.

How it works

TCS34725 RGB Sensor I²C · SDA/SCL Arduino Uno Classify Color threshold logic SG90 Servo Chute Position D9 PWM NEMA17 Stepper Rotate Bin Disk TB6600 Driver 20×4 LCD Live Counts 95% accuracy · hundreds of trials · ~10 items/min · under $106

What I did

I wrote all the Arduino firmware and wired the sensor and LCD. The sensor sends red, green, blue and clear readings over I²C, the Arduino checks them against threshold windows, a servo points the chute, and a stepper turns the bin disk. The LCD shows a running count per color. Most of my time went into the thresholds. Ambient light kept shifting the readings, so I ran around 100 test trials and kept adjusting until the machine sorted 95% correctly at about 10 items a minute. The whole build cost under $106.

What I learned

Sensor calibration is never finished; you keep measuring and adjusting. I also learned to keep firmware modular, one sketch per job, and to share grounds properly across a 3.3 V sensor, 5 V logic and a 12 V stepper. The machine won the class award for best accuracy and lowest cost.

FPGA binary conversion game

EC311 · Verilog · Xilinx Nexys A7 · 100 MHz · Jan-May 2025
Code
Nexys A7 board with the 7-segment display lit

The brief

Build an interactive system on a real FPGA in Verilog only, with no soft processor or firmware. It had to involve timing, state, input and a multiplexed display, all on a 100 MHz clock.

How it works

Random Counter LFSR / up-to 4095 Game FSM state control debounced BTN Binary→BCD 12-bit conversion integer division Display MUX time-division scan ≈1 kHz refresh 7-Seg LED 8 digits Nexys A7 SW[11:0] input LED_CORRECT output 100 MHz CLK · 4 testbenches · MIT license

What I built

The board shows a decimal number. The player enters it in binary on 12 switches and presses OK. A correct answer lights an LED and widens the number range. Under that is a top-level FSM, a binary-to-BCD converter, a 7-segment driver that multiplexes at about 1 kHz, button debounce logic and a clock divider. Each module has its own testbench in Vivado.

What I learned

Clock domains, propagation delay and synthesis constraints stopped being textbook ideas. Writing testbenches first made me think through edge cases before anything touched the board. It was the first time I took HDL from simulation to a bitstream running on real hardware.

Unit converter in WebAssembly

EC327 · C/C++ to WASM · Sep-Dec 2024
Try it Code

The brief

For EC327 (software engineering), a team of seven built a unit converter that works offline, covers 15+ categories, accepts custom unit rules and does its math in compiled code instead of plain JavaScript.

How it works

User Interface HTML + JS 15+ categories WASM Engine C / C++ → .wasm native precision Custom Units user-defined rules + / - / * / x History Log session record export .txt runs fully offline · no server required · all compute in-browser via WASM

What I did

I wrote back-end conversion modules in C/C++, compiled them to WebAssembly with Emscripten and handled integration testing, roughly half of the back-end work. The JavaScript front end sends inputs to the WASM module and keeps a history you can download. Users can also define their own units with a formula in x. A CORS problem blocked our local-server setup, so we packed everything into a single HTML file.

What I learned

How the WASM toolchain works end to end, how browser security rules can force an architecture change, and how to split front-end and back-end work cleanly in a big team.

YAGS branch predictor in gem5

EC513 · C++ · Jan-May 2026
Code

The problem

Global branch predictors suffer from aliasing: two different branches land on the same pattern history table entry and overwrite each other's history. Every resulting misprediction stalls the pipeline.

How it works

Branch PC XOR GHR Choice PHT 8192 × 2-bit ctrs bimodal bias T-Not-Taken Cache 8192 × {tag, ctr, v} T-Taken Cache 8192 × {tag, ctr, v} Prediction T / NT ~6% misprediction · 5 SPEC CPU 2017 benchmarks · compared against 5 gem5 predictors

What we built

We added YAGS (Yet Another Global Scheme) to gem5 as a new BPredUnit subclass. A choice table stores each branch's usual direction, and two tagged caches store only the exceptions. 13-bit tags from the branch address keep neighboring branches from colliding. We ran it on five SPEC CPU 2017 benchmarks (mcf, omnetpp, xalancbmk, gcc, x264) against five existing gem5 predictors and measured about 6% misprediction, close to the original paper.

What I learned

How a real simulator like gem5 is put together and how to plug a new hardware model into its pipeline. A segfault from an address overflow in the cache indexing taught me to bounds-check everything. The long simulations ran on BU's shared computing cluster, which meant learning to schedule jobs.

Team: Brian Quijada, Fadi Kidess, Ozan Ekame Pekgoz, Will Borland. Instructor: Prof. Ajay Joshi.

Self power-down SRAM sense amplifier

EC571 · Cadence Virtuoso · 50 nm · Jan-May 2026
Code

The problem

SRAM sense amplifiers trade speed against reliability. Latch-type amplifiers are fast and efficient but fail when the bitline swing is smaller than their offset, which gets worse at small nodes. Current-mirror amplifiers are reliable but keep drawing bias current the whole time sense-enable is on, long after the output has settled.

How it works

6T SRAM Cell BL / BLBAR VDD/2 precharge Level Shifter NMOS follower shift down ~Vt Current-Mirror SA PMOS mirror load 51 transistors APD Block Schmitt + NAND auto power-down PD Signal cuts bias off critical path 81% power reduction · 22.9 µW vs 123.6 µW · identical 3.1 ns delay · dual-V_HL for process corners

What we built

A current-mirror amplifier with a monitor circuit that notices when the output has settled and shuts off the bias current. The monitor uses Schmitt-trigger buffers, a high-skew NAND and a dynamic inverter. We added a second selectable Schmitt threshold to handle process variation across corners. Everything was designed and simulated in Cadence Virtuoso at 50 nm. Power dropped from 123.6 µW to 22.9 µW (81%) with the same 3.1 ns delay.

What I learned

Transistor-level design in an industry tool: sizing devices for target thresholds by hand calculation, designing for process corners, and cutting power without touching the critical path.

Team: Ozan Ekame Pekgoz, William Borland. Instructor: Prof. Rabia Yazicigil Kirby.