Skip to content

Navigation Menu

Sign in
Sign up
@abdulsamad2
abdulsamad2
Follow

Abdul Samad abdulsamad2

🎯
Focusing
Edge AI & TinyML — quantization and on-device benchmarking. MS researcher at Jeju National University (GKS). Six years full-stack before the lab.
  • Jeju, South Korea
  • 09:22 (UTC +09:00)

Block or report abdulsamad2

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
abdulsamad2 /README.md

Abdul Samad

Full-Stack Engineer · GKS Scholar · Edge AI & Applied ML Researcher

Six years shipping production web systems, now spending my research time on the harder half of the problem: getting machine learning off the notebook and onto real sensors, real devices, and real dashboards people actually use.

Email Upwork


About

Master's researcher in Electronics Engineering at Jeju National University, based in Jeju, South Korea on the Global Korea Scholarship. Before the lab, six years of full-stack engineering — APIs, dashboards, automation, and the unglamorous data plumbing that keeps them fed.

That combination is the point. Most edge-ML work fails somewhere between the sensor and the screen, and I have spent enough time on both ends to build the whole path rather than a slice of it.

Currently going deep on Edge AI and TinyML, computer vision, and time-series anomaly detection, applied to smart agriculture, environmental monitoring, energy management, and industrial safety.


What I Build

Systems Scalable full-stack applications, REST APIs, and real-time dashboards
Data High-volume synchronization pipelines and stream/batch processing
Ticketing Broker management platforms — inventory sync, pricing automation, margin reporting
Automation Web automation and resilient data-collection infrastructure
Vision YOLO-based detection for monitoring and safety use cases
Sensing Sensor-driven ML prototypes, from acquisition to inference
Delivery Edge and cloud deployment pipelines that survive contact with production

Ticketing & Resale Systems

A large part of my commercial work is custom software for ticket brokers — the operators reselling live event, sports, and concert inventory, where margin is won or lost on how fast and how accurately listings move.

Off-the-shelf tools handle the average broker. They do not handle yours. What I build instead is fitted to the operation:

  • Inventory management — one source of truth across every marketplace, with high-volume synchronization that keeps listings, splits, and holds accurate
  • Automated pricing — rules and data-driven repricing that react to demand and competitor movement instead of waiting on manual sweeps
  • Marketplace integrations — bulk listing, delisting, and order ingestion, so the same inventory works everywhere without duplicate effort
  • Operations dashboards — real-time visibility into sales, exposure, cost basis, and per-event margin
  • Automation — fulfillment, reconciliation, and reporting workflows that remove the repetitive manual work errors hide inside

The measure of success here is commercial, not technical: fewer oversells, faster turnaround, tighter pricing, and more profit per event.


Research Direction

Sensors → Real-Time Data → Machine Learning → Edge/Cloud Deployment → Decision Dashboard

Every stage in that chain is where projects break, so I work across all of them:

  • Edge AI and TinyML — inference under real memory and power budgets
  • Anomaly detection on noisy, drifting sensor streams
  • Computer vision for monitoring, inspection, and safety
  • Predictive irrigation and soil-moisture analysis
  • Agriculture and climate resilience
  • Energy monitoring and optimization

The goal is applied research that publishes and deploys — systems that keep working once they leave the laboratory.


Toolkit

AI & Data

Python PyTorch scikit-learn OpenCV YOLO Pandas NumPy

Full-Stack

TypeScript JavaScript React Next.js Node.js Express Tailwind CSS shadcn/ui

Data & Infrastructure

MongoDB PostgreSQL Redis Prisma Docker Linux Nginx PM2


Open To

Research collaborations, open-source work, internships, and industry–academia partnerships — particularly around edge AI and embedded ML, computer vision and sensor fusion, smart agriculture and climate tech, real-time data engineering, and AI-powered full-stack products.

📫 abdulsamad.laghari1@gmail.com · 💼 Upwork


Building reliable software today while researching intelligent systems for tomorrow.

Pinned Loading

  1. academyDashboard academyDashboard Public

    Admin dashboard for UHIL Academy, a Malaysian tutoring service — tutor matching, lesson scheduling, invoicing and deposits. Next.js 15, Prisma, NextAuth, Tailwind.

    TypeScript

  2. edge-ml-research-agent edge-ml-research-agent Public

    Human-in-the-loop research agent for edge/TinyML — CrewAI pipeline with approval gates, plus a Zephyr + Renode harness that benchmarks models on a simulated nRF52840 (exact flash/RAM from the linker).

    Jupyter Notebook

  3. fylo-landing-page fylo-landing-page Public

    tailwind css Project front end mentor challenge

    CSS

  4. JejuCentralMasjid JejuCentralMasjid Public

    Website for Jeju Central Masjid — prayer times, events and announcements, backed by Payload CMS. Next.js 16, TypeScript, Tailwind.

    TypeScript

  5. sensor-anomaly-baselines sensor-anomaly-baselines Public

    Anomaly detection baselines on 8 real NAB sensor streams — including a random control that beats every working detector under the common point-adjust protocol. Three evaluation protocols, honest up...

    Python

  6. tinyml-quantization-bench tinyml-quantization-bench Public

    Reproducible INT8-vs-float32 benchmark for TensorFlow Lite Micro on a simulated nRF52840 (Zephyr + Renode): 3.3x smaller, 1.8x faster, -0.1pp accuracy — with exact linker footprints and honest cave...

    C++

AltStyle によって変換されたページ (->オリジナル) /