Tuna Gül

I'm a third-year Civil Engineering student at Boğaziçi and I build things: AI agent systems, robots, and the software that holds them together. Most of my time right now goes into multi-agent systems. I wrote Blueprint, the tool that runs my own fleet of coding agents, and that fleet now does most of the work at Probot Studio, the robotics kit company I founded. Before agents, I wrote everything by hand, from a JSON parser in high school to firmware for competition robots.

I chose Civil Engineering at Boğaziçi on purpose. The people and the network here are worth more to me than what a CS degree somewhere else would teach me, and I have always learned software better by building it than in a classroom. I also love electronics and mechanics, so a lot of what follows ends up as a physical thing.

LinkedIn · Email · CV (PDF)

Tuna working on a competition robot, wiring its electronics.

AI

All AI →

Machines that learn and the societies they form: AI agents that work as one team, LLM agents in economic games, and the hackathon builds I directed with agent teams.

blueprint (bp) Live

A command-line tool that turns AI coding agents into one team that can find and message each other, on one machine or across many.

Role: AuthorPeriod: 2026–

Ridgeline of smoothed choice distributions in the 11-20 money request game: humans (ochre, front) mostly ask for 17 or 18, while LLM agents under seven persona methods (navy rows) pile up at 19 and 20.

LLM agents in economic games Research

Putting LLM agents into real economics experiments, alone or next to human players, to see what kind of society they form when they share a pot, punish and vote on their own rules.

Illustration of a light swarm of persona dots dropping answers, one column at a time, into a stacked-dot response distribution that fills up to, but never above, a plain ochre baseline curve.

simengine Research

A library that simulates how a population responds, using LLM personas, and checks every result against simple baselines.

Illustration of a weekly AI lesson: two white cards, on the left a small 2-4-4-1 neural network with one path from input to output picked out in dark navy, on the right its learning curve falling over training, with twelve evenly spaced checkpoint dots and the last one in ochre.

AI Lead, CompEC

I lead AI at CompEC, Boğaziçi University's computer club, and give weekly AI lessons to students.

Robotics

All robotics →

Probot Studio, where we make robot kits and the tools to learn with them, and the FIRST robotics teams I have co-founded, programmed for and mentored since high school.

Probot Studio

We design, make and sell modular metal robot kits, and build the tools to learn with them. Design in Builder, code in Blocks, drive in Sim, then build it with real parts.

A robot made in Probot Builder — Drive in Sim ↗ Open in Builder ↗

Probot Studio Live

We make robot kits and the tools to learn with them, so a student can go from a first robot to competition level. Part of the İTÜ Çekirdek Acceleration Program.

Role: Founder · software, sales, strategyPeriod: 2023–

▶ Play

Probot Blocks Live

Block coding that turns into C++ for our ESP32 library or into FTC Java.

▶ Play

Probot Builder Live

A 3D robot builder in the browser. Put a robot together from our real kit parts, get its parts list, and check it against FTC rules before you buy anything.

▶ Play

Probot Sim Live

A robot simulator in the browser that runs students' FTC Blocks and Java programs, with a multiplayer match server.

probot-core Live

A source-available ESP32 library for competition robots, free for education and competitions: the robot hosts a Driver Station in the browser over Wi-Fi.

A simulated differential-drive robot starts about 0.6 m off a curved reference path and converges onto it under RAMSETE path following, with robot footprints along its trail; an inset card shows the wheel velocity step response with feedforward alone settling low and PID plus feedforward reaching the setpoint.

probot-sdk Live

A header-only C++ control stack for ESP32 robots: PID, LQR, Kalman filtering, odometry and a command scheduler.

FIRST robotics

The whole story →

I co-founded FRC 7839 in high school, then after graduating we started 9029. Now I lead FTC Zağanos.

Seven simulated shooter trajectories from different distances arc up and fall through the mouth of a raised goal basket; the highlighted shot leaves a small robot's turret with backspin and is shown at equal time steps. The physics is computed; the robot and goal are stylised.

FTC Zağanos #25577 Won

FIRST Tech Challenge team. FTC Türkiye Champions, the Control Award, and the Reach Award at the FIRST Championship in Houston.

FRC 9029 Team NF Won

The FIRST Robotics Competition team we started after high school. 2025 Ankara Regional winner.

Illustration of a vision coprocessor: a camera with an LED ring wired to a small screen, where a half-hexagon reflective target glows in a dark frame with its detected contour, corner points and a centroid crosshair.

NightVision (FRC 7839) Archive

The vision coprocessor of FRC 7839, the team I co-founded in high school: a terminal UI on a small screen, an Arduino control panel, and the link to the robot.

Organising robot competitions

I co-organise the impact.tr Hands-On Challenge robot competitions in Istanbul with a friend.

Other robots

Robot work outside FIRST and Probot Studio.

Illustration of a floor-cleaning robot's coverage plan: an occupancy-grid map of an L-shaped room with a round table and furniture, long horizontal back-and-forth passes joined by rounded U-turns, the cleaned strip behind the robot and the dashed passes still ahead; a small dock sits on the left wall.

Floor-cleaning robot (ROS 2) Archive

A coverage path planner for a ROS 2 floor-cleaning robot: from a SLAM map to parallel sweep passes.

Systems

All systems →

Embedded, parsers, data structures and algorithms, written by hand.

Illustration of the method with simulated readings: six Hall sensors on a ring, a distance circle around each, the pairwise intersections nearest the magnet's path, and their average marking the magnet and the shaft angle.

Hall-sensor absolute encoder Archive

An attempt at a cheap absolute encoder for our FRC robot. A ring of Hall sensors around a magnet, and the circle geometry to find the magnet, worked out by hand in C.

Role: Algorithm and C codePeriod: Jul–Oct 2022

Building cross-section with four elevator shafts, the cars at their floors and passengers queued at the landings, fed by a small floor-versus-time trace of the last minute, from a small dispatch simulation.

Elevator dispatch simulator Won

A building and elevator simulator in Rust, written in a hackathon week with one teammate, with cascaded PID control for every car. Our team took 1st place at Boğaziçi AI UpForAChallenge.

Two-camera distance to targets Archive

Distance to the reflective FRC targets from two cheap cameras, without a full stereo calibration. Blobs are paired across the two frames, and distance is fitted against 1/disparity from real measurements.

Illustration of a streaming JSON decoder: a small nested list drawn as an indented tree on a white card, its deeper part boxed in amber as a lazy generator below the max_depth cut, leaving the card through a port beside the root bracket as a stream of token chips on a winding track, with chevrons showing the direction, that runs off to the right.

DBEX Archive

A JSON-compatible encoder and decoder in Python built on generators, so it can stop at a chosen depth and hand back the rest lazily. On PyPI since 11th grade.

All projects →

Now

Running my AI agent fleet on blueprint, giving weekly AI lessons as AI Lead at CompEC, building Probot Studio in the İTÜ Çekirdek Acceleration Program, and leading FTC Zağanos as team lead for the BIOBUZZ season.

Updated: 2026-10-23

News

2026-08 Probot Studio became part of the İTÜ Çekirdek Acceleration Program.
2026-05 FTC Zağanos won the Reach Award at the FIRST Championship in Houston.
2025-12 FTC Zağanos became FTC Türkiye Champions and won the Control Award.