THE JOURNEY

SOFTWARE · SENSING · ROBOTICS

From code to
the physical world.

From backend applications to sensing research, I keep following the connection between information and action.

Follow the thread
ALI ZARGARI · 2014—NOWWORK / RESEARCH / EXPERIMENTSJump into the tactile lab

2014–2020

Built backend APIs and taught coding.

I wrote my first Hello World in 2014. At CASA in 2015–2016, I began building backend applications and APIs. Bakery work from 2017–2020 brought a different kind of troubleshooting: keeping the shop’s computers working. Teaching children at Code Ninjas in 2019–2020 made clear explanations part of the job. Software was becoming a way to help someone do something.

2021–2023

Automated RF measurements with Python.

At Element Materials Technologies, I wrote Python scripts that communicated with RF test equipment and parsed its data. The result depended on understanding the instrument as well as the code. I began software engineering at San José State University in 2022, bringing that practical experience into coursework and team projects.

2024

Built software for recipes, social cues, and memory.

For DishSocial, I helped connect the team’s database, backend, and recipe-sharing interface. SocialSync moved that integration toward assistance: I helped build a responsive camera-inference workflow for facial-expression feedback. It was a working prototype, without established clinical outcomes. For Memento, a team memory-assistance project, I built the original FastAPI/PostgreSQL backend. Each project asked how captured information could become useful to someone.

Late 2024

Added Kalman localization and LiDAR obstacle avoidance.

In our ROS2/Webots vehicle simulation, I implemented a Kalman-filter node and added LiDAR obstacle avoidance to camera lane following. Noisy readings now affected the next action. I graduated from SJSU in December 2024, having been named an Engineering Dean’s Scholar in 2023, with a growing interest in the connection between sensing and control.

2025–2026

Added reconnects, clocks, and replay to ReverseChess.

ReverseChess pushed me beyond game rules into repeated move requests, reconnects, clocks, and replay. RepoCoach applied the same attention to repository analysis: background jobs, stored evidence, and reports someone could return to. Both are working prototypes. Their less visible parts became some of the most interesting work—keeping state consistent and making the next step understandable.

That thread continues in the Personal Atlas Project, an ongoing native macOS application. I connected encrypted personal-history imports and source-linked search to on-device inference, then added experimental adapter training with reviewed datasets, checkpoints, and separate evaluation. The workflow lets someone inspect where an answer came from and what the model used.

2024–present

Built sensing and device pipelines for MindMesh.

Olympus began as a more responsive home and grew into MindMesh, an ongoing ambient-intelligence project for elderly care, sleep monitoring, and home automation. I work across mechanical, electrical, and software questions: mmWave sensing, embedded devices, communication, and simulation. Firmware and signal-processing pipelines are implemented; broader orchestration is ongoing. I continued this work alongside my work at SERES in summer 2026.

2026

Led tactile ML and automated sensor testing.

I began master’s studies in Robotics and Automation at Santa Clara University in March. From June through August, I led tactile ML and model development at SERES to improve contact-position and force estimation for a dexterous robot finger. I designed the test bed and built force control, automated tests, graphing, and analysis. I also researched and developed reference-point selection, spatial-resolution, test-matrix, and calibration methodologies.

SERES / REPORTED RESEARCH

FULL-SCALE ERROR<1%

Improved model · research setup

MODEL FIT≈0.99 R²

Evaluated with MAE and RMSE

Two training runs and one holdout run, with millions of data points at each stage.

MANUAL → AUTOMATED COLLECTION

2–3Millions

Complete touch-location-and-force measurements per minute.

SERES results, reported separately from the synthetic browser experiment. The live lab displays its own measured errors.

TRY THE EXPERIMENT

Train a model to estimate contact and force.

Teach a small network to read a synthetic magnetic fingertip.

Magnetic tactile sensing: contact, magnetic displacement, and measurementA cutaway diagram of a compliant sensing surface. A probe indents the elastomer, shifting embedded magnets relative to a fixed sensor board. Magnetic measurements feed the contact-estimation model. This explains the principle and is not a measured research result.APPLIED FORCEELASTOMERMAGNETSSENSORSINFERENCE
System diagram: contact deforms the sensing surface; a learned model estimates position and force.
What spatial resolution measures
CALIBRATION COVERAGE
Where training examples were collected. This lab samples contacts across the surface, plus no-contact examples; the displayed grid illustrates coverage.
OUTPUT REFERENCE POINTS
Which location an output describes. This lab predicts one contact’s X, Y, and normal force relative to a defined surface.
SPATIAL RESOLUTION
How close two contact locations can be and still be reliably distinguished. Measuring this requires controlled tests and an error criterion; grid density or low average error alone does not establish it.
Related work: ReSkin, Bhirangi et al. ↗

Now

Continuing robotics study and sensing projects.

I’m continuing graduate study while developing MindMesh and Personal Atlas. My next focus is connecting learning systems to reliable physical behavior, from the measurement behind a prediction to the action it informs. There is more to build, test, and understand.

CURRENT WORK

Continuing robotics research
in sensing and control.

Working on sensing, robotics, or software that connects to the physical world? I’d like to hear about it.

Let’s compare notes

Page and tactile lab created with Astra. My engineering work, with project teams credited in the story.