AI & Software Engineer · Generative AI · Sensor AI · Applied ML · Embedded Systems

Building intelligent systems — from embedded software and Sensor AI to modern Generative AI.

I am an AI and software engineer with experience across embedded systems, deep learning, and applied research. My work spans sensor intelligence, time-series modelling, and intelligent software development, alongside my growing focus on large language models, retrieval-augmented generation, and agentic AI applications.

Also building in GenAI · LLM agents, RAG & GraphRAG
Explore TFT-IS: Measured Spectra, EMG Fits & PCA ↗ Explore selected research Google Scholar
MSCA doctoral researcher at Saarland University · Open to industrial AI, machine learning, Sensor AI, and software engineering opportunities.
2019+continuous research in sensor AI
8accepted/published journal articles across sensor AI
7first-author journal articles
5+years in product engineering
Research identity

One sensor-intelligence framework across sensing modalities.

The sensing modality changes, but the research logic remains consistent: understand the measurement, preserve informative structure, learn representations that generalise, and design for practical acquisition and deployment.

01 Measure

Acquire informative IMU time series or impedance spectra.

02 Model

Expose temporal, spectral, and physical signal structure.

03 Learn

Build compact, transferable, and interpretable representations.

04 Deploy

Reduce sensing, computation, and instrumentation cost.

Applied across two sensing modalities

Wearable inertial signals describe human behaviour; impedance spectra describe chemical interactions. Both are treated as structured physical measurements rather than generic input data.

Wearable and physical signals

Human motion through wearable inertial sensors

Since 2019, I have studied how body-worn accelerometers and gyroscopes can support reliable human-centred intelligence, spanning activity recognition, gait-based affective computing, cross-domain representation transfer, and sleep monitoring.

  • Human activity recognition from accelerometer and gyroscope time series
  • Transfer of motion representations to data-scarce emotion tasks
  • Wearable AIoT systems for sleep–wake monitoring
  • Compact CNN, recurrent, and Transformer-based models for edge use
IEEE IoT Journal ↗ IEEE Sensors Journal ↗
Wearable motion intelligence
Wearable inertial sensing workflow Accelerometer and gyroscope waveforms flow into a learned motion representation used by activity, emotion, and sleep tasks. AXAYAZGYRO IMU windows z Task heads Activity Emotion Sleep
Motion signals→Reusable representations→Human-centred tasks
Chemical and electrical signals

Chemical state through impedance spectroscopy

My doctoral work studies how frequency-dependent complex impedance can reveal richer chemical information than a single resistance measurement, while remaining compatible with compact and cost-aware measurement systems.

  • Interpretable full-spectrum impedance models and compact features
  • Single-gas and mixture analysis
  • Efficient strategies
Read preprint ↗ Dataset & code ↗
Frequency-domain sensor response
Illustrative impedance response curves Three smooth curves represent sensor impedance measured at different operating temperatures. log frequencyimaginary impedance 350 °C250 °C150 °C
Dense spectra→Model-derived features→Evalution
Methods shared across modalities
Time-series modellingRepresentation learningExperimental validationResource-aware inferenceEmbedded deployment
Representative work

Different signals, one research programme.

01 · Cross-domain transfer

From Steps to Sentiments

Transfers representations learned from human activity recognition to activity-based emotion detection, connecting large-scale motion learning with a smaller affective-computing task.

Transfer learningWearable IMUsEmotion
IEEE Internet of Things Journal ↗
02 · Wearable AIoT

Inertial sensing for sleep monitoring

Combines wearable motion data with a Transformer–CNN model for practical sleep–wake monitoring, extending inertial intelligence beyond conventional activity labels.

Sleep–wakeTransformer–CNNAIoT
IEEE Embedded Systems Letters ↗
03 · Efficient wearable ML

Resource-conscious human activity recognition

Develops compact CNN and CNN–BiGRU pipelines for recognising everyday and sports activities from inertial signals while retaining a deployment-oriented model design.

HARCNN–BiGRUEdge AI
IEEE Sensors Journal ↗
04 · Pseudo-supervised learning

Learning food spoilage stages without external labels

Transforms structure discovered in temperature-cycled MOS sensor responses into temporally ordered pseudo-labels, enabling supervised classification of Fresh, Mildly Spoiled, and Spoiled food states without external chemical or microbiological ground-truth labels.

Pseudo-supervised MLMOS sensorsFood spoilage
IEEE Sensors Letters · Accepted
05 · Impedance spectroscopy

Temperature–frequency-tuned impedance spectroscopy

Models frequency-dependent impedance spectra using an exponentially modified Gaussian (EMG), extracting four interpretable spectral features—amplitude, position, width, and asymmetry—to analyze concentration-dependent responses, distinguish gases, and investigate gas mixtures across temperatures.

EISInterpretable featuresMOS sensorsOpen data
Research Square preprint ↗
Selected publications

A publication record spanning human-centred wearable intelligence and frequency-domain chemical sensing.

Full record on Scholar ↗
Current doctoral direction

Chemical sensing, food quality, and impedance spectroscopy

2026
IEEE Sensors LettersPublishedFood spoilage sensing

A pseudo-supervised approach for food spoilage detection using Metal Oxide Semiconductor gas sensors and temperature modulation

Introduces a pseudo-supervised learning framework that derives temporally ordered Fresh, Mildly Spoiled, and Spoiled labels from patterns discovered in MOS sensor responses, enabling supervised food-spoilage classification without external chemical or microbiological ground-truth labels.

IEEE Xplore ↗
↗
2026
Research SquarePreprintGas sensing

Temperature–frequency-tuned impedance spectroscopy with interpretable full-spectrum modelling enables robust gas sensing

Introduces a temperature- and frequency-resolved framework for compact, interpretable modelling of MOS-sensor impedance in single gases and mixtures.

↗
Established research track

Inertial and wearable sensing

2026
IEEE Internet of Things JournalTransfer learningWearables

From Steps to Sentiments: Cross-Domain Transfer Learning for Activity-Based Emotion Detection in Wearable IoT Systems

Transfers movement representations from activity recognition to emotion classification using wearable inertial sensing.

PublisherPDF
↗
2026
IEEE Embedded Systems LettersSleep monitoringAIoT

Artificial Intelligence of Things–Enabled Sleep Monitoring Using Wearable Inertial Sensors and a Transformer–CNN Model

Combines inertial actigraphy with a compact Transformer–CNN architecture for practical sleep–wake monitoring.

PublisherPDF
↗
2024
SensorsHuman motion analysisDeep learning

Generisch-Net: A Generic Deep Model for Analyzing Human Motion with Wearable Sensors in the Internet of Health Things

Introduces a general-purpose deep learning framework for analysing human motion from wearable inertial sensors in Internet of Health Things applications.

Publisher
↗
2023
IEEE Sensors JournalHuman activity recognitionDeep learning

Smart-Wearable Sensors and CNN-BiGRU Model: A Powerful Combination for Human Activity Recognition

Develops a wearable-sensor recognition pipeline that combines local temporal feature extraction with bidirectional sequence modelling.

PublisherPDF
↗
2023
Applied SciencesAffective computingGait

Machines Perceive Emotions: Identifying Affective States from Human Gait Using On-Body Smart Devices

Studies emotion recognition from motion captured by body-worn devices, connecting gait analysis with affect-aware intelligent systems.

PublisherPDF
↗
2022
IEEE Sensors LettersSingle-authorEfficient HAR

Khail-Net: A Shallow Convolutional Neural Network for Recognizing Sports Activities Using Wearable Inertial Sensors

Explores a shallow, resource-conscious convolutional design for sports-activity recognition from wearable inertial signals.

PublisherPDF
↗
2022
IEEE Sensors LettersSingle-authorHuman activity recognition

UltaNet: An Antithesis Neural Network for Recognizing Human Activity Using Inertial Sensors Signals

Proposes a GRU-first, convolution-second architecture for recognising human activities from wearable inertial signals, reversing the more common CNN–recurrent processing order.

Publisher
↗
Experience

Research shaped by sensor AI, measurement science, and product engineering.

My academic work applies a consistent sensor-intelligence approach across wearable inertial signals and impedance-based gas sensing, grounded in more than five years of embedded, automation, and machine-learning engineering experience.

2024 — present

MSCA Doctoral Researcher

Saarland University · SERENADE

Doctoral research in Systems Engineering on intelligent MOS sensor systems, spanning excitation strategies, impedance modelling, unsupervised and pseudo-supervised learning, food-state monitoring, and data-driven sensor representations. The programme includes industrial research periods with Bosch Sensortec.

Host: Lab for Measurement Technology, Saarland University Industry collaboration: Bosch Sensortec GmbH
Dec 2018 — Apr 2024

Engineering roles at Emumba

Emumba Private Limited

Progressed through embedded-systems, quality-automation, and generative-AI roles over more than five years.

ML Engineer (Generative AI)Dec 2023 — Apr 2024

Developed LLM-based assistants and retrieval-augmented generation pipelines for backend applications.

QA EngineerJan 2023 — Nov 2023

Built Python-based GUI test automation, supported technical hiring, and trained engineers in Python, Linux, and Git.

Design Engineer (Embedded Systems)Dec 2018 — Dec 2022

Developed C/C++ firmware, custom embedded-Linux and Yocto images, and rapid IoT prototypes for MCUs and single-board computers.

Completed degrees
MS, Computer Science National University of Sciences and Technology - School of Electrical Engineering and Computer Science · Islamabad, Pakistan
BS, Electrical EngineeringNational University of Computer and Emerging Sciences · Islamabad, Pakistan
Practice projects

Applied AI systems built to test design choices end to end.

Independent projects using fictional data to practise system design, evaluation, and technical communication.

01 · Completed baseline Traditional RAG

FakeCompanyABC: evaluated retrieval before generation

A fully local RAG pipeline over fictional HR policies, developed through chunking experiments, dense retrieval, policy routing, reranking, grounded generation, and evaluator comparison.

100 benchmark questions 1.000 Hit@3 ~67 ms selected CPU retrieval
Nomic embeddingsChromaDBTF-IDF routingQwen
View project details GitHub ↗
02 · Work in progress GraphRAG

FakeCompanyABC: constructing a validated policy graph

An extension of the same corpus toward GraphRAG, progressing from entity-schema discovery and GLiNER validation into text-grounded relation-candidate discovery before relation-schema validation and Neo4j evaluation.

25 entity types 361 stored detections 161 adjacent relation candidates
GLiNEREntity validationRelation discoveryNeo4j
View project details Entity viewer ↗ Relation viewer ↗
03 · Evaluated · In progress Agentic AI

GharMind: agentic AI and evidence-guided self-correction

Previously HomeMind — about the rename ↗

Independent local agentic-AI research with Qwen3, LangChain, Ollama and simulated smart-home controls. Current Stage 02.4 studies evidence-guided self-correction: a 0.6B model extracts claims, Python verifies execution evidence, and the 1.7B main LLM revises unsupported answers without extra tool access. On 100 scenarios: 76 full passes, 17 partial, 7 failures (Stage 02.3: 75/15/10). This is a preliminary between-run comparison, not proof of causal improvement.

02.4 self-correction study 76 / 100 qualitative full passes 33 / 35 verified write sequences 100 isolated test scenarios
Qwen3 1.7B + 0.6BEvidence-guided correctionPython verificationOllamaLangChainLLM-as-judge
Explore self-correction results GitHub ↗
04 · Implemented Edge-RAG

Portfolio Edge-RAG: validated retrieval with browser-local generation

A static-site RAG chatbot built from 13 curated portfolio sources and 88 heading-aware chunks. A 100-question benchmark compared Dense MiniLM, BM25, and RRF; BM25 was selected, then paired with compact evidence selection, Qwen3-0.6B browser generation, and an extractive fallback. Generation is manually smoke-tested; detailed judge-based generation evaluation remains a future validation task.

100 grounded questions 94% BM25 Hit@3 88 portfolio chunks
BM25MiniLM benchmarkQwen3-0.6BTransformers.jsJudge evaluation planned

Demo notice: The hosted chatbot may fail on GitHub Pages. The locally served version works; cloning and run instructions are on the project page.

View project details
Technical blogs · 2019–2020

Blogs & engineering notes.

Practical articles from my earlier work on embedded Linux, connected devices, and development toolchains — a foundation for my later work in applied AI and intelligent systems.

All articles on Medium ↗
01 / 05FEB 2020 · CLOUD & IoT

End-to-End IoT with AWS IoT Core

Raspberry Pi sensing, ESP8266, MQTT, a Yocto integration, and Grafana monitoring.

Read article ↗ 02 / 05NOV 2019 · EMBEDDED LINUX

Python Packages in Yocto Images

Integrating Python packages into custom Linux images for embedded devices.

Read article ↗ 03 / 05OCT 2019 · VOICE & IoT

Raspi-Echo: A Raspberry Pi Voice Assistant

Building an Alexa-based voice assistant on Raspberry Pi using the Alexa SDK.

Read article ↗ 04 / 05AUG 2019 · SYSTEMS ENGINEERING

Why & How to Create a Custom Toolchain

A practical introduction to custom toolchains for embedded development.

Read article ↗ 05 / 05AUG 2019 · CROSS-COMPILATION

Cross-Compilation for Banana Pi

Building software for an ARM-based target from a separate development machine.

Read article ↗

Selected historical technical posts; original publication dates and articles are preserved on Medium. Browse the full archive ↗

Teaching & online courses

Practical learning resources for deep learning and Linux.

Beyond research and engineering, I create free, hands-on courses that help learners develop practical skills in neural networks and open-source tools.

Explore the courses

Learn by building, step by step.

01 · Deep learningFree Udemy course

From Zero to Hero: Mastering Neural Networks with Keras

A hands-on introduction to training neural networks with Keras and Google Colab, including convolutional, recurrent, and hybrid models for image and human-activity recognition.

KerasGoogle ColabCNNLSTM
★ 4.7/524 ratings
2,997students
1h 20m5 lectures
Explore course on Udemy
02 · Linux fundamentalsFree Udemy course

From Novice to Ninja: Linux Basics for Everyone

A beginner-friendly guide to installing Linux, setting up virtual machines, navigating files and directories, using essential shell commands, and managing software packages.

LinuxCLIVirtualBoxPackage management
★ 3.9/566 ratings
3,745students
45m16 lectures
Explore course on Udemy

Udemy course statistics checked 10 October 2026. The ≈4.1/5 combined rating is an estimate weighted by each course's number of ratings, not an official instructor rating. Enrollments are summed across courses and may not reflect unique students. Udemy instructor profile ↗

Open and reproducible work

Research artefacts that can be inspected, rerun, and extended.

The TFT-IS repository contains dense impedance measurements, curated single-gas and gas-mixture datasets, feature-extraction pipelines, and browser-ready notebooks.

Open repository ↗
01
Gas-mixture workflow

Reproduce EMG feature extraction and PCA analysis for ethanol–acetone mixtures.

↗ 02
Single-gas workflow

Reproduce first-sweep feature extraction and pairwise PCA analyses.

↗ 03
Code portfolio

Browse research repositories spanning sensing, wearable intelligence, and engineering tools.

↗
About

Building intelligent systems — from embedded software to applied AI.

I am an AI and software engineer with an MSc in Computer Science and a background spanning embedded systems, machine learning, and applied research. Currently, I am a Marie Skłodowska-Curie doctoral researcher at Saarland University, developing intelligent sensing methods that combine experimental measurements, signal processing, and machine learning.

My engineering experience includes embedded Linux, C/C++, Python, IoT systems, and software development. My research has focused on deep learning for wearable inertial sensors and intelligent chemical sensing, with an emphasis on robust models, real-world validation, and efficient deployment.

More recently, I have been expanding my work into Generative AI, including large language models, retrieval-augmented generation (RAG), GraphRAG, and agentic systems. I am particularly interested in bringing these technologies together to develop reliable, practical AI applications.

Core capabilities

Applied AI & Machine Learning

Deep learning · Time-series modelling · Transfer learning · Representation learning · Model evaluation

Generative AI & LLMs

LLMs · RAG · GraphRAG · Agentic AI · Tool calling · Prompt engineering

Software & Embedded Systems

Python · C/C++ · Embedded Linux · IoT · Firmware · Software integration

Research & Intelligent Sensing

Wearable sensors · Human activity recognition · Impedance spectroscopy · Signal processing · Experimental validation

Contact

Interested in Generative AI, Sensor AI, or intelligent software systems?

I welcome opportunities in industrial AI, machine learning, and software engineering, as well as collaborations on Generative AI, Sensor AI, and applied research.

engrhamzaaliimran@gmail.com