Skip to content
View ahmedsohail2003's full-sized avatar
  • Waterloo, Ontario, Canada

Block or report ahmedsohail2003

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.

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
ahmedsohail2003/README.md

Ahmed Sohail Butt

Computer Science student at Wilfrid Laurier University building practical systems across robotics & embodied AI, data engineering, cloud infrastructure, and payments.

I enjoy the part of software delivery where a loosely defined problem becomes clear requirements, measurable success criteria, and working software.

🌐 Portfolio: ahmedsohail2003.github.io β€” the five-project WorkCell series on one page, with demos.

πŸ€– Physical AI: one simulated industrial work-cell, five learning approaches

One robot (SO-ARM100 in MuJoCo), one manipulation task, taught five different ways β€” with honest measurement throughout. Every project below shares the same simulated work-cell, publishes its datasets/models, and reports failures as carefully as successes.

WorkCell series Approach Headline result
Part A Β· Sim2Cell Imitation learning (ACT) + domain randomization 90% success, 87–100% robust under randomization β€” incl. finding & correcting a false "0% collapse" caused by an eval-harness rendering bug + a claim-by-claim ACT-paper reproduction report
Part B Β· TalkToTheCell Vision-language-action (SmolVLA 450M fine-tune, free-tier GPU) 0% β†’ 90%, commanded in natural language; the same dataset moved ACT βˆ’15 and SmolVLA +35 β€” a two-sided architecture-vs-data study
Part C Β· GraspSight Classical perception: RGB-D β†’ point cloud β†’ 6-DoF pose β†’ grasp 20/20 end-to-end picks, 0.58 mm ICP pose error, 0.32 mm hand-eye calibration
Part D Β· PolicyForge Reinforcement learning + from-scratch world model World-model planner solves Reach in 8k env steps (37Γ— fewer than PPO); contact-rich Lift solved via a 4-stage reward-design study
Part E Β· CellOps ROS 2: SLAM + Nav2, C++ grasp service 12/12 Nav2 goals on a self-built SLAM map; Part C's estimator ported to C++ at 0.000 mm parity, 3.3Γ— faster, served over ROS 2
Datasets & models 4 published LeRobot datasets + 5 model cards on πŸ€— Every result above is reproducible from public artifacts

Upstream open source: two open pull requests to LeRobot, currently under maintainer review β€” #4174 (Windows Hub path normalization) and #4177 (safe video-backend selection) β€” plus independent Windows verification of #4121.

Featured work

An end-to-end data product for a residential painting business:

  • 22 functional requirements, user flows, and measurable acceptance criteria
  • Python ETL with pandas, SQLAlchemy, watchdog, and automated tests
  • Azure SQL provisioned through Bicep and Azure CLI
  • Typed SQL reporting views and a .NET Blazor/Dapper analytics dashboard
  • Privacy-safe, fixed-seed synthetic demo data

A four-person academic project comparing symbolic and generative AI agents:

  • Simple, Minimax, Alpha-Beta, and Gemini-powered players
  • My ownership: Gemini prompt construction, response validation, corrective re-prompting, and evaluation
  • Metrics for search nodes, real API requests, first-try validity, and retries
  • Python test suite and GitHub Actions CI

An educational e-Transfer-style system built as two Spring Boot microservices:

  • REST-based account and transfer services with separate data ownership
  • Account, amount, currency, and insufficient-funds validation
  • Saga-style compensation when a credit fails after a debit
  • 45 JUnit, Mockito, and MockMvc tests
  • OpenAPI, Docker Compose, Kubernetes, and CI

Tools I work with

Languages: Python, SQL, Java, C#/.NET, JavaScript, C/C++

Robotics & ML: PyTorch, LeRobot, Hugging Face, MuJoCo, RL (PPO/SAC), world models, VLA fine-tuning (SmolVLA), 6-DoF grasping, sim-to-real methodology

AI and data: Google GenAI SDK, prompt engineering, pandas, NumPy, SQLAlchemy, ETL pipelines, statistical evaluation

Cloud and backend: Azure SQL, Bicep, Azure CLI, Spring Boot, REST APIs, Dapper, Docker, Kubernetes

Delivery: Requirements analysis, user flows, acceptance criteria, technical documentation, relational modelling, Git/GitHub

Currently focused on

  • Building AI prototypes with observable quality and explicit fallback behaviour
  • Designing trustworthy data workflows from raw ingestion through decision-ready reporting
  • Learning distributed-systems and payments concepts through testable projects

Contact

Popular repositories Loading

  1. Multi-Agent-Adversarial-Game Multi-Agent-Adversarial-Game Public

    Benchmarks Minimax and Alpha-Beta agents against a Google Gemini agent with validation, corrective re-prompting, and evaluation metrics.

    Python 1

  2. YPYW-Business-Intelligence YPYW-Business-Intelligence Public

    Full-stack BI platform with Python ETL, Azure SQL IaC, analytics views, and a .NET Blazor dashboard built from documented business requirements.

    HTML

  3. transfer-platform transfer-platform Public

    Educational e-Transfer-style platform using Spring Boot microservices, REST orchestration, compensation logic, Docker, Kubernetes, and 45 tests.

    Java

  4. ahmedsohail2003 ahmedsohail2003 Public

    GitHub profile README for Ahmed Sohail Butt β€” AI, data, cloud, and payments projects.

  5. workcell-partA-imitation workcell-partA-imitation Public

    End-to-end imitation learning on a MuJoCo SO-ARM100 work-cell: scripted expert -> LeRobotDataset -> ACT -> domain-randomization robustness study (90% nominal, 87-100% under DR)

    Python

  6. workcell-partC-grasping workcell-partC-grasping Public

    From-scratch RGB-D perception-to-grasp pipeline in MuJoCo: segmentation -> point cloud -> 6-DoF pose (yaw-sweep + trimmed ICP, 0.58mm) -> grasp -> closed-loop pick (20/20)

    Python