Muhammad Ahmed - Data Scientist, AI Engineer and Kaggler

01 About me

From research papers
to Kaggle podiums.

introduction

I'm a Data Scientist, AI Engineer and Kaggle Competitions Master with a passion for unlocking the power of data. I specialize in computer vision, machine learning and natural language processing, and I've delivered results in both research projects and Kaggle competitions.

My research covers self-supervised video understanding: aligning human activity in time and segmenting actions without labels, using video transformers and optimal transport. On Kaggle I've medalled twenty times, in problems as different as protein engineering, LLM-generated text detection, reinforcement learning and gravitational waves.

I hold a Bachelor's degree in Computer Science from FAST NUCES, Karachi, and I'm always open to new collaborations in data science and AI.

Download CV Get in touch

kaggle.com/muhammad4hmed

Pakistan's first and only Kaggle Competitions Master

3 gold
8 silver
9 bronze

currently

AI Engineer, Pioneer Corporation

Tokyo, Japan · remote · since October 2023

previously Research Engineer (CV & ML) at Retrocausal · Data Scientist at Pikky

{
  "role": "AI Research Engineer",
  "focus": ["vision", "language", "LLMs"],
  "kaggle": {
    "tier": "Competitions Master",
    "gold": 3, "silver": 8, "bronze": 9
  },
  "papers": ["ECCV", "WACV", "IEEE ICCWS"],
  "based_in": "Karachi, Pakistan",
  "open_to": "collaborations"
}

/blog · published daily

The Daily AI Brief

What actually shipped in AI: models, papers, benchmarks and tooling, with the details that matter.

Technical competence

Data Science & AI

Machine Learning
Deep Learning
Computer Vision
Natural Language Processing
Data Analysis

Frameworks

PyTorch
TensorFlow
Scikit-learn
Keras
OpenCV
Pandas
NumPy

Programming & Fundamentals

Python
C++
SQL
Data Structures and Algorithms

Containerization & Cloud

Docker
Amazon Web Services
On-click services

Soft Skills

English Communication
Problem-solving
Teamwork
Critical Thinking
Adaptability
Business Acumen

03 Experience

Academic and professional career

Oct 2023 — Present

AI Engineer

Pioneer Corporation

Tokyo, Japan Remote
Jul 2022 — Oct 2023

Research Engineer - Computer Vision and Machine Learning

Retrocausal

Redmond, WA, USA Remote
Dec 2021 — Jun 2022

Data Scientist

Pikky

India Remote
Nov 2020 — Dec 2021

Software Engineer (Intern), Machine Learning Infrastructure

Retrocausal

Redmond, WA, USA Remote
2018 — 2022

education

Bachelor of Science, Computer Science

FAST National University of Computer and Emerging Sciences

Karachi, Pakistan

04 Achievements

Accomplishments in competitions

2023

Won a Kaggle competition solo

2022

Became Pakistan's first and only Kaggle Competitions Master

2020

Became Kaggle Competitions Expert, the 3rd and youngest in Pakistan to reach the rank

Kaggle leaderboard finishes

20 medals across computer vision, NLP, LLMs, reinforcement learning, signal processing and 3D reconstruction.

Enzyme Stability Prediction SoloProtein engineering rank 2 of 2,482 teams top 0.08%
LLM - Detect AI Generated TextLarge language models rank 3 of 4,358 teams top 0.07%
U.S. Patent Phrase to Phrase MatchingPhrase similarity prediction using NLP transformers rank 10 of 1,889 teams top 0.53%
Hungry GeeseGame bot using reinforcement learning rank 19 of 879 teams top 2.2%
CommonLit - Evaluate Student SummariesText score prediction using LLMs rank 27 of 2,064 teams top 1.3%
Image Matching Challenge 2024 - Hexathlon3D scene reconstruction from 2D images across six domains rank 34 of 929 teams top 3.7%
G2Net Gravitational Wave DetectionSignals from black-hole collisions using signal processing rank 41 of 1,219 teams top 3.4%
Feedback Prize - Evaluating Student WritingArgumentative writing analysis using NLP transformers rank 52 of 2,058 teams top 2.5%
CommonLit Readability PrizePassage complexity prediction using NLP transformers rank 60 of 3,633 teams top 1.7%
Cassava Leaf Disease ClassificationImage classification using computer vision rank 94 of 3,900 teams top 2.4%
ROGII - Wellbore Geology PredictionGeology prediction for automated oil and gas drilling rank 248 of 6,125 teams top 4.0%
Google Landmark Recognition 2020Landmark recognition in images using computer vision rank 62 of 736 teams top 8.4%
Jigsaw Multilingual Toxic Comment ClassificationBilingual NLP transformers rank 86 of 1,621 teams top 5.3%
SIIM-FISABIO-RSNA COVID-19 DetectionCOVID-19 detection from chest X-ray and CT images rank 127 of 1,305 teams top 9.7%
Shopee - Price Match GuaranteeUnsupervised multimodal product matching from text, images and metadata rank 128 of 2,426 teams top 5.3%
OpenVaccine: COVID-19 mRNA Vaccine DegradationRNA degradation prediction using NLP rank 141 of 1,636 teams top 8.6%
PetFinder.my - Pawpularity ContestPhoto popularity from images and metadata using CV and NLP rank 186 of 3,537 teams top 5.3%
Riiid! Answer Correctness PredictionStudent knowledge tracing with time-series models rank 288 of 3,406 teams top 8.5%
Mechanisms of Action (MoA) PredictionData science & ML, improved an existing algorithm's accuracy rank 316 of 4,373 teams top 7.2%
Orbit WarsReal-time strategy game agent in continuous 2D space rank 380 of 4,729 teams top 8.0%

Publications

ECCV 2024
Learning by Aligning 2D Skeleton Sequences and Multi-Modality Fusion
Paper · Demo video
WACV 2024
Permutation-Aware Activity Segmentation via Unsupervised Frame-to-Segment Alignment
Paper · Demo video
IEEE ICCWS 2022
Sequential Embedding-based Attentive (SEA) classifier for malware classification
Published at IEEE ICCWS 2022

Other competitions

2ndInternational Data Analytics Olympiad (IDAO) 2022, qualifiersAmong teams from Stanford, IIT, HSE University, the University of Tokyo and others; qualified for the World Finals in Moscow, Russia, representing my country and university
2ndGeoffrey Hinton Hackathon (Data Science & ML)Against professionals and students from top Indian universities such as IIT and NIT
1stMachine Learning Competition by HackerEarthOut of 5,164 participants
1stData Science Competition by Dphi.tech1st position
2ndData Science Competition by Dphi.tech2nd rank
6thHackerEarth Deep Learning CompetitionOut of 5,494 participants
5thMelanoma Tumor Size Prediction by MachineHack - Analytics IndiaOut of 300 participants
5thJanataHack: Demand Forecasting by Analytics VidhyaOut of 300+ participants
9thJanataHack: Customer Segmentation by Analytics VidhyaOut of 500+ participants
6thJanataHack: Machine Learning in Agriculture by Analytics VidhyaOut of 600+ teams
RU·1ACM League (Competitive Programming) by ACM MAJU1st runner-up
8thICPC Topi RegionalRanked 8th
RU·2Procom'19 & DevDay'19 (Speed Programming) by FAST NUCES Karachi2nd runner-up
1stCodeWars 2018 (Competitive Programming) by ACM - FAST NUCESWinners

05 Projects

Research and open source

The Cadre canvas: four agents joined by delegate and hand-off arrows

VS Code extension
Multi-agent
MIT

Cadre — build a team of AI agents in VS Code

Draw the agents and the arrows between them, hit launch, and watch them all work at once, one live lane per agent. A delegate arrow makes one agent a tool of another; a hand-off arrow passes a finished result along. Describe the pipeline in a sentence and Claude draws it for you. Runs on the Claude Code subscription you already have, with no API keys.

Marketplace Code

Diagram of the LA2DS skeleton-alignment framework

ECCV 2024
Self-supervised
Video transformer

Learning by Aligning 2D Skeleton Sequences and Multi-Modality Fusion

We propose a novel self-supervised temporal video alignment framework for fine-grained human activity understanding. By using 2D skeleton heatmaps instead of 3D coordinates, our approach leverages spatial and temporal self-attention in a video transformer, achieving higher accuracy and robustness.

PDF Video

WACV 2024
Unsupervised
Optimal transport

Permutation-Aware Activity Segmentation via Unsupervised Frame-to-Segment Alignment

Our paper presents a novel transformer-based framework for unsupervised activity segmentation, incorporating both frame-level and segment-level cues. By leveraging a frame-level prediction module and unsupervised training via temporal optimal transport, along with a segment-level prediction module and frame-to-segment alignment module, our approach achieves permutation-aware segmentation results.

PDF Video

Open source
AutoML
Python

GML - Auto Data Science

Tired of doing Data Science manually? GML is here for you! GML is an automatic data science library in python built on top of multiple Python packages.

Code

Architecture of the SEA malware classifier

IEEE ICCWS 2022
NLP
Security

Sequential Embedding-based Attentive (SEA) classifier for malware classification

The rise in smart devices has brought about increased security risks, including the threat of malware. Detecting and combating malware early on is crucial to prevent widespread device corruption and network failure. Our research introduces a lightweight, efficient malware detection model utilizing state-of-the-art NLP techniques.

PDF Code

✦ Book a call

Do you have any questions?

Schedule a meeting and let's discuss!

Send a request

06 References

Testimonials

I had a pleasure of working with Muhammad at the Retrocausal, collaborating on couple of projects. I was impressed Muhammad's execution ability under tight timeline, he is very good at delivery results with the minimum resources. He would be a great asset to any team!

Andrey KoninChief Architect at Retrocausal

Ahmed is extremely good with modelling irrespective of size of dataset. It was a pleasure to work with him at Pikky and follow the thought process he used to come up with the recommendation model, among other things.

Alaknanda AgarwalSenior Product Manager, ICICI Lombard

07 Contact me

Get in touch

A project, a research idea or a role in mind? Send a message, or book a call.

Full name

Email

Subject

Message