Things I’ve built

Helps organizations show their AI is managed responsibly: AI systems, risks, policies, sign-offs and evidence in one workflow, checked against recognized standards, with human-approved AI review and audit-ready reports.

My role

Platform design, development & deployment, end to end

• I built the core workflow that keeps AI systems, risk assessments, policies, sign-offs, lifecycle checkpoints and evidence together and turns them into audit-ready reports.
• I built the AI-assisted review that reads documents, evidence and connected code repositories, shows its progress as it works, and waits for a person’s approval before any verdict is used.
• I added automated quality checks that test the AI’s judgment against known answers before any change to how it reviews is accepted.
• I built the security side: strong sign-in, encrypted secrets, role-based access, each organization’s data kept separate, and protection against manipulated inputs.
• I added a tamper-evident record of every action, and deletions that need approval from another person, for audits and accountability.

Results & status

Research platform · 2026

Built the workflow from AI system inventory and risk assessment through sign-offs, supporting evidence, and audit-ready reports.

Delivered assisted review with visible progress, human approval, and quality checks against known answers before review changes are accepted.

A team of specialized AI agents that carries a research question and its papers to neuro-inspired model ideas and working code, on locally hosted models, tracing and evaluating its own work.

My role

Backend & frontend, part of the infrastructure; model monitoring & evaluation

• I developed the system’s backend and its frontend structure.
• I worked on part of the system’s infrastructure.
• I set up monitoring of the AI models, tracing each agent’s work so every result can be inspected.
• I evaluated the models’ performance and worked on improving it, with automatic evaluation in which AI reviewers score the system’s outputs.
• I also worked on improving the agents.

Results & status

Research platform · 2026

Developed the backend and frontend structure and contributed to the infrastructure supporting the research workflow.

Added monitoring and automated output evaluation, then worked on improvements to the models and agents.

A research health platform connecting participant enrollment, mobile recordings, and results from the team’s AI models.

My role

Platform design, development & deployment

• I designed, built, and deployed the research platform.
• I integrated the team’s models and built the UbiCell and UbiCell Kid companion apps.

Results & status

Deployed research platform · 2026

Delivered a shared platform for recording uploads, model processing, and participant results.

The Android study app for D-Dream: biometric login, guided face or fingertip recording, fail-safe local storage that syncs once online, and an estimated white blood cell count with a history of past results.

My role

Android app design & development

• I designed the whole app.
• I built biometric login, enrollment, and face or fingertip video capture for D-Dream studies.
• I built the step-by-step recording flow that helps a participant hold a fingertip steady on the camera with the flash on.
• I made data capture fail-safe: each camera recording is saved to local storage, waits in a queue and syncs with the D-Dream server once the phone has internet.
• I built the results screen, with the estimated white blood cell count, its reference range and a clear note that it is not a medical diagnosis, plus a history grouped by day and week.

Results & status

Android research app · 2026

Built enrollment, biometric login, guided face or fingertip recording, and the local storage and upload queue.

Delivered an estimated white blood cell result screen with a reference range, a non-diagnostic label, and daily and weekly history.

A child-friendly UbiCell, built in collaboration with Children’s Wisconsin: each fingertip recording becomes a mission, with animations, themes and a mini game that hold a child’s attention until the mission report.

My role

Android app design & development

• I designed the whole app for children: instead of a plain video collection, each measurement is a mission, from the start of the recording to the mission report.
• I built the animations, themes and mini game that attract and hold a child’s attention while the fingertip video records.
• I made data capture fail-safe: each camera recording is saved to local storage, waits in a queue and syncs with the server once the phone has internet.
• I added biometric login.
• I built the app on the same D-Dream backend as UbiCell, so both apps share one server.

Results & status

Android research app · 2026

Designed and built the mission-based recording experience and result report for the collaboration with Children’s Wisconsin.

Implemented biometric login, local recording storage, and queued uploads to the backend shared with UbiCell.

A suicide prevention research platform for participant enrollment, consent, surveys, and ongoing communication.

My role

Sole developer: design, backend, frontend & deployment

• I designed, built, and deployed the full platform as sole developer.
• I added two-way messaging, gift cards, and automatic study-data transfers.

Results & status

Deployed research platform · 2026

Delivered one platform for study participation, follow-up, and day-to-day coordination.

A suicide prevention research platform for participant surveys, automated messages, and study reports.

My role

Sole developer: design, backend, frontend & deployment

• I designed, built, and deployed the full platform as sole developer.
• I created the surveys, public forms, automated messages, and response-based reports.

Results & status

Deployed research platform · 2026

Delivered survey collection with downloadable study reports and daily email reporting.

A web and Android platform for collecting and tracking participant data across eight study profiles.

My role

Sole developer: portal, server & Android app, design to deployment

• I designed, built, and deployed the portal and Android app as sole developer.
• I added coordinator roles, collection tracking, bulk downloads, and an audit log.

Results & status

Deployed research platform · 2026

Delivered eight condition-specific collection profiles for field research.

The Amadergram field app: biometric login, a participant found by QR code, surveys, vitals, documents and face or fingertip video under fixed conditions, and a verified upload queue that keeps data on the phone if it fails.

My role

Android app design & development

• I designed the app from scratch and built it in Kotlin with AI-assisted development, including biometric login.
• I built the QR code scanner that fetches a specific participant’s data before collection starts (lookup by participant ID also works), with a completeness view across demographics, vital signs, lab results and videos.
• I built face and fingertip video recording under fixed conditions of distance, position and lighting.
• I built survey data collection and the capture screens for demographics and documents, with configurable recordings for heart rate, HRV, blood pressure, SpO2 and respiration rate.
• I built the upload flow: a queue whose records the server verifies, with local storage that keeps the data on the phone when an upload fails.
• I set up the Django server on an AWS EC2 instance and the AWS S3 bucket that receive the collected data.

Results & status

Android research app · 2026

Built biometric login, participant lookup, a completeness view, and capture screens for the study’s different data types.

Delivered face and fingertip recording under fixed capture conditions and a verified upload queue connected to the study server and AWS S3.

Product website introducing UbiRing, a companion app for smart rings and bracelets.

My role

Website design & automated deployment

• I designed the UbiRing brand, logo, and website.
• I built and launched the product page with scroll animations and automated publishing.

Results & status

Product website · 2026

Delivered the product website and a consistent visual identity for UbiRing.

A companion app for iOS and Android that pairs with smart rings and bracelets, uploads their readings for the lab’s research and turns them into dashboards: vitals, a night timeline with sleep stages and readiness scores.

My role

App design & development for iOS & Android

• I built the app from scratch, including its full design, with React for both iOS and Android and with AI-assisted development.
• I built device pairing with smart rings and bracelets on top of the vendor SDKs; the SDK for the ring hardware itself was provided.
• I built the vitals screens for heart rate, blood oxygen, HRV and stress, with live readings and spot checks.
• I added many dashboard visualizations, including a night timeline with sleep stages and readiness views such as body battery and stamina.
• I built the upload of measurements to the server for the lab’s neuro-physiological monitoring research.

Results & status

iOS and Android research app · 2026

Designed and built the iOS and Android app, including pairing with smart rings and bracelets and uploading measurements to the server.

Delivered live and spot-check vitals, sleep-stage timelines, and readiness views. The portfolio collage uses demo data.

A self-hosted observability and evaluation layer for the lab’s research agents: it traces each request and agent step, tracks latency and tokens, gathers them in dashboards and judges generated ideas with LLM-as-a-judge.

My role

LLMOps & observability: deployment, integration, dashboards & evaluation

• I deployed self-hosted Langfuse with automated deployment.
• I connected it to the other microservices of the agent infrastructure, so everything the agents do is tracked.
• I set up tracing for each request and each agent’s steps, with latency and token-consumption tracking.
• I created the dashboards, and set up Prometheus and Grafana for operational monitoring of the services.
• I wrote LLM-as-a-judge prompts for different types of judge to evaluate the generated ideas, run by local Nemotron evaluators.

Results & status

Research infrastructure · 2026

Deployed and integrated self-hosted tracing, request and agent-step tracking, latency and token dashboards, and operational monitoring.

Wrote prompts for different types of judge and connected local evaluators to assess the ideas generated by the research agents.

Builds on the open-source MODNet portrait-matting project to remove or replace the background behind a person in images, videos and live camera feeds, locally on NVIDIA GPU, Apple GPU or CPU. 140+ GitHub stars.

My role

Developer, extending open-source MODNet

• I built on the existing open-source MODNet project and made it easier to use.
• I added video support and background-image replacement.
• I improved the user experience and built a Flask application on top of the model workflow.
• I added execution options for NVIDIA GPU (CUDA), Apple GPU and CPU.
• I packaged the tool as CPU and CUDA Docker images and wrote tests with Pytest.

Results & status

Open-source tool

Delivered background removal and replacement for still images, video files, and live camera input.

Added a Flask application, CPU and CUDA Docker packaging, and Pytest coverage around the model workflow.

A Cefalo hackathon project: an LLM-based tool that runs the social media posting workflow as one pipeline, from an input job through AI steps on a local LLM to scheduling, with no manual step.

My role

AI pipeline & backend

• I built the AI pipeline around a local LLM.
• I built the backend structure that runs the full pipeline, from the input job to scheduling.
• I architected the system so the posting workflow runs end to end with no manual step.

Results & status

Hackathon project · 2025

Built the AI pipeline and backend structure carrying jobs through the posting workflow to scheduling.

Connected the pipeline stages into the hackathon’s automated workflow. My contribution covered the AI and backend; the interface was another contributor’s work.

A personal study project, a warm-up before a hackathon: asynchronous Python topic publishers and subscribers on Azure Service Bus, tested with messages from 1 MB to 100 MB.

My role

Developer (personal study)

• I built this test project on my own, as a warm-up before a hackathon, to learn how Azure Service Bus handles event messaging.
• I wrote asynchronous topic publishers and subscribers in Python with asyncio.
• I implemented two authentication paths: connection strings and passwordless authentication.
• I tested the implementation on the Premium tier with message sizes from 1 MB to 100 MB, and packaged it to run in Docker.

Results & status

Open-source study project · 2024

Published asynchronous topic publisher and subscriber examples with both authentication options.

Packaged the implementation in Docker and ran Premium-tier message-size feasibility tests. This is separate from the employer’s messaging migration.

Detection and segmentation of small objects (drones and birds in the sky, aerial-image features and underwater fish), each task with its own model, trained on augmented datasets and served through a web API.

My role

Developer (training assignment)

• I chose the model family that fits each task.
• I prepared the flying-object, aerial-image and NOAA underwater fish datasets and augmented them with color saturation changes, random crops and image flips.
• I trained and evaluated the models, using DarkNet, ResNet-101 and Inception-v2 for feature extraction and TensorBoard to report on training. The drone detector reached 74% accuracy.
• I built a Flask web API that serves the models’ predictions.

Results & status

Training project · 2020

Built YOLOv3 drone detection, UNet aerial-image segmentation, and Faster R-CNN fish detection workflows.

Delivered dataset preparation, augmentation, model training and evaluation, TensorBoard reporting, and a Flask prediction API.

An open-source collection of practical computer vision demos with Python, OpenCV and MediaPipe: hand detection, face mesh, Haar-cascade face and eye detection, image inpainting and image-to-text.

My role

Developer

• I built hand detection and face mesh demos with MediaPipe.
• I implemented face and eye detection with OpenCV Haar cascades.
• I put together image inpainting experiments.
• I wrote image-to-text examples.

Results & status

Open-source examples · 2022

Published hand-detection and face-mesh demonstrations alongside Haar-cascade face and eye detection examples.

Added image-inpainting experiments and image-to-text examples to the same open-source collection.

An open-source collection that turns hard-to-read data, from COVID-19 figures and call logs to Bangladesh bank deposit rates, into interactive Streamlit dashboards and Jupyter notebooks with Plotly and Matplotlib.

My role

Developer

• I built Streamlit dashboards for COVID-19 analysis, call log analysis and Bangladesh bank deposit rates.
• I created visual analysis notebooks in Jupyter with Plotly and Matplotlib.
• I published the project site at mazhar004.github.io/python-data-visualization alongside the source code.

Results & status

Open-source dashboard collection · 2019–2026

Built dashboards for COVID-19 analysis, call logs, and Bangladesh bank deposit rates.

Published the notebooks and Plotly/Matplotlib visualizations with their source code.

Two Django web apps on external APIs, published as one site: WebLocate shows where the server behind a website or IP address is, and Similar Movie suggests related movies and series with ratings and trailers.

My role

Developer

• I built both apps in Django and connected them to external APIs for location lookup and movie data.
• I built WebLocate’s lookup for website and IP addresses, including the “my” shortcut for the user’s own server information.
• I built Similar Movie’s suggestions of related movies and series, showing ratings, trailers and extra details.
• I designed the interfaces with HTML, CSS and Bootstrap and published both apps as one web app on PythonAnywhere.

Results & status

Personal web project · 2019

Built website and IP-address location lookup, including the “my” shortcut for the user’s own server information.

Built related movie and television suggestions with ratings and trailers, and published the combined application on PythonAnywhere in 2019.

An internal tool for managing lab inventory, with access tailored to each user’s role.

My role

Web application development & deployment

• I built the inventory management app for the research lab.
• I added role-based access and automated deployment.

Results & status

Deployed internal tool · 2026

Deployed an internal inventory tool with access tailored to lab members’ responsibilities.

The public website of Marquette University’s ubiquitous-computing laboratory at ubicomp.cs.mu.edu, where outside visitors learn about the lab’s research areas, people and publications.

My role

Website design, development & deployment

• I designed the site’s look and layout.
• I built the pages for research areas, people and publications.
• I deployed the site and put it live at ubicomp.cs.mu.edu.

Results & status

Public lab website · 2026

Designed the visual layout and built the research, people, and publication pages.

Deployed the site at the lab’s university address, giving the research group one public home for its work and members.

Company website introducing Ubitrix and its businesses in compliance, wearables, and remote health.

My role

Website design, development & deployment

• I designed and built the company website and its page layouts.
• I launched it on the company domain with secure hosting and automated publishing.

Results & status

Company website · 2026

Delivered a public website connecting Ubitrix’s three sister companies.