TRAFFIC-COP · FRAUD DETECTION IN PRODUCTION SINCE 2022REVENUE CLAWBACKS 5% → 0.52%

Hamza Faheem Mir

Senior ML Engineer · Data Scientist

I build production AI systems that turn messy data into measurable revenue.

Revenue recovered annually
$5M+
IVT clawbacks cut 5% → 0.52%
Reduction in case resolution time
80%
GPT-powered Ad-Ops dashboards
Compute speedup
50×
75 hours → 1.5 hours via PySpark
MAPE improvement
18%
Prophet revenue forecasting
Hamza Faheem Mir
About Me

Turning data chaos into production AI

Senior ML Engineer with 5+ years building end-to-end ML pipelines, fraud-detection systems, and GPT-powered analytics tools in Python, PySpark, Airflow, and Kubernetes on AWS. I've recovered ~$5M/yr in ad revenue, cut manual workflows by 80%, and mentored engineering teams, bridging the gap between data science research and production reality.

Islamabad, Pakistan
Education
B.S. Computer Science
National University of Computer & Emerging Sciences (NUCES)
2015 – 2019 · Data Mining · Natural Language Processing · Artificial Intelligence · Bio-informatics
Off Duty
Sat Pakistan's civil-service exam: International Relations, Sociology, US History, Criminology. Otherwise, football.
Technical Skills

The stack behind the results

ML & AI

PythonMachine Learningscikit-learnLangChainProphetMLflowStreamlitFlask

Data & Orchestration

PySparkApache SparkAirflowDatabricksSQLAmazon EMR

MLOps & Cloud

DockerKubernetesAWSAWS S3AWS AthenaSageMakerCI/CD

Analytics & Visualization

Metabasepandasnumpyseabornstatsmodels

Tools & Auth

GitBitbucketAWS CognitoRoute 53
Work Experience

Where I've made an impact

Jan 2022 – Present
Remote

Senior Machine Learning Engineer

MonetizeMore · Canada-based AdTech · Google Certified Publishing Partner
  • Led the "Traffic Cop" invalid-traffic programme using Airflow + PySpark on AWS (S3 + Athena); reduced revenue clawbacks from 5% to 0.52%, saving ~$5M/yr.
  • Shipped two GPT-powered analytics apps (Website Analysis Agent & IVT Dashboard) on Kubernetes with Flask, Streamlit, and LangChain, adopted by 100% of Ad-Ops staff and cutting manual case-resolution time 60–80%.
  • Engineered a Prophet-based revenue-forecast pipeline with logistic-growth modelling; improved MAPE by 18% across two products via variant benchmarking and Airflow-scheduled retrains.
  • Mentored three junior engineers to full autonomy; codified ML code-review guidelines and best-practice playbooks to raise team-wide code quality.
Apr 2021 – Dec 2021
Remote

Data Scientist

Baltoro · ML/AI data platform · US-based
  • Delivered geospatial service-time predictions in PySpark MLlib for a transportation client; parallelised across Spark executors to cut compute time 50× (75h → 1.5h).
  • Optimised probabilistic tennis-match models with MapReduce; improved training throughput and reliability at scale.
  • Prototyped a trigger-based ML execution feature enabling on-demand benchmarking and reproducible reruns.
Projects

Systems I've built and shipped

ShippedSub-minute audit latency

Website Analysis Agent

Point it at a website and get a full site-health and ad-compliance audit in under 60 seconds.

Kubernetes-native Flask API running a Playwright scraper, piping JavaScript signals to a LangChain LLM, and rendering insights via Streamlit, delivering on-demand site-health and ad-compliance audits.

LangChainFlaskStreamlitKubernetesPlaywrightLLM
Shipped80% time reduction

IVT Dashboard with LangChain

A GPT dashboard that answers Ad-Ops questions and cut validation time by 80%.

GPT-powered Ad-Ops dashboard (Flask + Streamlit + LangChain) deployed on Kubernetes, secured with AWS Cognito authentication and exposed globally via Route 53. Cut validation time by 80%.

LangChainFlaskStreamlitKubernetesAWS CognitoRoute 53
ShippedProduction IVT scoring

Risk-Level Assessment Pipeline

Scores invalid traffic across the whole network hierarchy and streams results to live dashboards.

PySpark + Airflow DAG scoring invalid traffic across the Internet Network Hierarchy; results persisted to S3, auto-hydrating Athena tables for real-time Metabase monitoring.

PySparkAirflowAWS S3AthenaMetabaseMLlib
Shipped18% MAPE improvement

Revenue Forecasting with Prophet

Predicts product revenue 18% more accurately and retrains itself on schedule.

Airflow-orchestrated Prophet pipeline improving MAPE by 18% across two products via logistic-growth modelling with cap/floor guardrails and automated scheduled retrains.

ProphetAirflowPythonForecasting
ShippedEnd-to-end MLOps

MLOps Rules-Generation Pipeline

Turns decision-tree models into production rules and ships them automatically.

End-to-end SageMaker + Airflow pipeline that derives decision-tree rules, registers models in MLflow, and ships them via Airflow-driven CI/CD to AWS S3/Athena.

SageMakerAirflowDockerCI/CDAWSMLflow
ShippedMulti-source unification

Centralised API Data Integration

Pulls Stripe, Google Ads, and survey data into one real-time dataset for analytics.

Standardised and unified data ingestion from Stripe, Google Ads/Ad Manager, and SurveySparrow via Airflow ETL pipelines, enabling real-time consolidated datasets for analytics and dashboarding.

AirflowETLStripeGoogle AdsPython
Operator Log

Beyond the model

[LOG 01]

Cross-Team Range

  • Worked across three teams: ML, Data Engineering, and the Traffic Cop product team
  • The bridge between ML and Data Engineering, representing ML in cross-team work
  • Solely owned MLOps for the ML team: Kubernetes deployments, Cognito authentication, Route 53 domains
[LOG 02]

Team Leadership

  • Mentored junior engineers and data scientists
  • Coordinated ML, Data Engineering, and product stakeholders
  • Ran sprint planning and technical roadmaps
[LOG 03]

Product Thinking

  • Owned AI features end-to-end, from user research to metrics to analytics dashboards
  • Translated business requirements into ML system specs
  • Defined success metrics before writing a single model
  • Shipped features adopted by 100% of Ad-Ops staff
[LOG 04]

Production Scale

  • Designed for high-throughput, low-latency serving
  • Automated pipelines that cut manual workflows by 80%
  • Built on AWS with observability from day one
Get In Touch

Let's build something
great together

I'm currently open to new roles, freelance projects, and consulting opportunities. Whether you have a project in mind or just want to say hello, my inbox is open.

Ready to see my full resume?
Download the PDF for the complete picture: experience, projects, and skills.
DOWNLOAD RESUME