Master thesis•Prosthetics & rehabilitation

Biosignals & Prosthetics Engineer

SmitSakariya

I turn muscle signals into hand movement. I build the sensor chain, the pattern recognition and a virtual prosthetic hand that patients can practise with at home, before their real prosthesis is ready.

Portrait of Smit Sakariya
Berlin · M.Sc. AIMS
  1. Muscleforearm sEMG + FMG
  2. Front-endamplify · ESP32
  3. Filter20–450 Hz · 50 Hz notch
  4. FeaturesMAV · RMS · WL
  5. ClassifierLDA per user
  6. Handvirtual prosthesis

Profile

About

From the clinic to the code: I have installed medical devices in hospitals, and now I build the software that helps patients learn to use a prosthesis.

I am an M.Sc. student in Analytical Instruments, Measurement and Sensor Technology at Coburg University, with a bachelor's degree in Biomedical Engineering from Gujarat Technological University.

For my bachelor project I built a 3D-printed prosthetic arm controlled by EMG. For my master thesis, planned with Think3DDD / 3D-Medico in Berlin, I am building a virtual prosthesis trainer: an EMG + FMG band and a camera drive a virtual hand on the patient's scanned limb, so training can start at home.

Before my master's I spent 15 months with hospitals and diagnostic labs as a medical device engineer, installing devices, training clinical staff and fixing problems on site.

4hand states decoded from muscle signals: rest, open, close, pinch
8grip types detected from the camera hand model
16pressure sensors (FSR) in the simulated FMG band
<150 mstarget end-to-end latency, muscle to virtual hand

Selected work

3 projects
Project 01 · Master thesis

Virtual Prosthesis Trainer In progress

2026 · Coburg University · planned with Think3DDD / 3D-Medico, Berlin

A myoelectric prosthesis is hard to learn, and training usually starts only after the socket is made. This system lets a patient practise earlier, at home, with a virtual prosthetic hand mounted on their own limb scan.

  • Real-time pipeline: band signals → filtering → features → per-user LDA classifier → virtual hand over WebSocket
  • Camera hand tracking copies every finger and the wrist, and labels training data automatically
  • Detects 8 grip types (power, pinch, tripod, key grip, point and more) from 3D hand geometry
  • One Euro filtering removes jitter without adding lag
  • Grasp-and-place game logs success rate, grab time and release time for every session
  • Records and replays hand movements for therapist review
PythonsEMG / FMGLDAMediaPipeThree.jsWebSocket
GameplayRecorded with camera tracking · blue band = sensor band position
Game 1 · 1:43Build a towerGrab blocks and stack them. Round 2: pick each block up from the table yourself.
Game 2 · 1:31Fruit basketClose the hand around the fruit, carry it, open over the basket. Round 2: reach for it yourself.
Game 3 · 2:31Match itPlace each object on the matching colour, then shape, then colour + shape, with a score card at the end.
3D model of the sensor band on a forearm: 8 printed pods with 16 pressure sensors, 2 EMG sensors and the ESP32 box
16 FSR in 8 printed pods · 2 EMG · ESP32 box

The sensor band

A 3D-printed band worn on the forearm below the elbow. 16 pressure sensors (FMG) feel how the muscles change shape, 2 EMG sensors read the electrical activity of the finger flexors and extensors, and a wireless ESP32 sends everything to the trainer.

  1. Wear the band. The patient puts the band on the residual limb and opens the trainer on a laptop.
  2. Teach the AI. The patient tries a fist, a pinch, an open hand, a few seconds each. A small classifier learns this person's own muscle pattern.
  3. Move the virtual hand. The band streams wirelessly, the AI decodes the intended grip, and the virtual prosthetic hand on the patient's own limb scan opens and closes in real time.
  4. Train with games. Fruit basket, tower and matching games practise grasp, carry and release, and log success rate and grab and release time, so the therapist sees progress before the real socket is ready.

Today a camera and a blue marker band stand in for the sensors (videos above); the printed band replaces them next.

Open the 3D band ↗
Project 02 · Bachelor project

EMG prosthetic hand

Medical engineering · Gujarat Technological University, Ahmedabad

A tendon-driven, 3D-printed prosthetic hand controlled by forearm muscle signals. Flex the muscle, and the hand closes.

  • Designed and 3D-printed an articulated hand with jointed fingers, thumb and palm, with tendons pulled by servo motors
  • Surface EMG sensor on the forearm; an Arduino turns muscle activity into finger motion
  • Integrated EMG sensor and electrodes, Arduino, servos and battery into one portable unit
  • Filtered and processed the noisy raw muscle signal (OpenViBE and code) into a clean control input
sEMGArduinoServo / tendon driveOpenViBE3D printing
Prosthetic hand connected to a laptop and EMG electronics
Prosthetic hand gripping a bell pepper
Arduino, wiring and battery inside the forearm shell
Project 03 · Medical imaging

Tuberculosis chest X-ray detector

Personal project · 2026

An object detector that marks tuberculosis regions on chest X-rays. I trained it on the public TBX11K dataset to learn the full workflow for medical image data: preparing labels, training, and judging a model with the right metric.

  • YOLOv8 trained in Google Colab on TBX11K
  • Evaluated with mAP at IoU 0.5 on held-out images
  • Same skills used in the thesis: dataset handling, validation, honest reporting
YOLOv8PyTorchMedical imagingColab
0.345
mAP50 · TB detection · TBX11K
Scale 0 to 1 · first baseline model
Other relevant work

UWB localization

Sensor project · ESP32 + Decawave DW3000

Finds the position of a moving tag from its distances to fixed anchor modules, using ultra-wideband radio.

  • ESP32-DW3000 modules as fixed anchors and one tag, connected over SPI
  • Distance by two-way ranging (time of flight), which cancels the clock offset between modules
  • Position by trilateration; an extra anchor adds redundancy and lowers the error
  • Logged distance and signal strength per anchor, plotted the tag with an uncertainty circle
ESP32UWBTrilaterationPython

Crash Pirates

3D game · browser and Godot

A 3D car-combat game: ram rivals, grab weapons from mystery boxes and steal the crown.

  • Real-time 3D in the browser with Three.js, ported to the Godot engine for Android and Windows
  • Bot AI, weapon effects, touch controls and player ranks
  • The same game skills build the therapy games of the prosthesis trainer
Three.jsGodotGame AI

Outmeet

Android app · live on Google Play

Find, join and host in-person sports, fitness and social events nearby.

  • Flutter app with a Node.js and Socket.io real-time backend, live map and group chats
  • Published on Google Play with sign-in, push notifications and privacy pages
  • Shows I can ship an app real users install at home, like the patient trainer
FlutterNode.jsSocket.io

Experience

Work & education
Sep 2025 – Mar 2026
Berlin, Germany

Intern, App Development (AI-assisted)

Think3DDD GmbH
  • Built the front-end of a 3D scanning mobile app (similar to KIRI Engine) using AI-assisted development: directing AI coding tools to generate and iterate the app code, then reviewing and testing it
  • The app captured video scans of patients' hands and legs for prosthetic doctors and fed them to a backend pipeline that turned them into 3D models for 3D-printed prostheses
  • Owned the integration layer between the backend processing software and the front-end app, for a smooth data flow from scan capture to 3D model output
  • Reviewed, tested and refined the generated front-end code to meet real clinical usability requirements
  • Linked the backend engineering team with non-technical clinical users, turning clinical and UX feedback into concrete implementation changes
  • Focused on a responsive, easy-to-use interface for prosthetic doctors without a technical background
May 2023 – Aug 2024
Ahmedabad, India

Medical Device Sales & Service Engineer

Jainam Biomedical
  • Installed laboratory devices (hematology and biochemistry analyzers, coagulometers) in hospitals, clinics and diagnostic centres
  • Trained doctors and lab technicians on device use
  • Troubleshooting, preventive maintenance and after-sales technical support
  • Product demonstrations, market analysis and B2B sales across Gujarat
Education
Oct 2024 – present
Coburg, Germany

M.Sc. Analytical Instruments, Measurement & Sensor Technology

Coburg University of Applied Sciences
Jul 2019 – Jun 2023
Ahmedabad, India

B.E. Biomedical Engineering

Gujarat Technological University
Skills
Biosignals

Surface EMG · FMG · filtering · feature extraction (MAV, RMS, WL, ZC) · LDA / SVM · calibration

Software

Python · JavaScript · Three.js · MediaPipe · WebSocket · YOLOv8

Hardware

Arduino · ESP32 · sensors · 3D printing · Blender

Clinical & languages

Device installation · staff training · service
English (fluent) · German (B1) · Gujarati, Hindi (native)

Let's build better prosthetic training together.

smitgsakariya01@gmail.com

Berlin, Germany · open to master thesis and working student positions