Moeen Bagheri

About Me

I’m an AI Engineer with 12+ years of professional experience in AI engineering and software development. I hold a Master of Science in Data Science and Analytics from Ryerson University. At BASF I lead FOX Talk2Data, an AI assistant used by 4,000+ people that turns questions into SQL and explains financial results. The model writes the narrative, and a check rejects any figure the data does not support. I also built a coding agent that takes a task through tests and pull-request review.

Email LinkedIn GitHub Toronto, ON, Canada

Skills & Expertise

Concepts

Agentic Coding Multi-Agent Orchestration AI Agent Design & Development Agent Tool Design Fullstack & API Development Evaluation & Testing CI/CD RAG Time-Series Forecasting Finetuning Transfer Learning Computer Vision NLP

Soft Skills

Technical Leadership & Mentorship Stakeholder Communication

Languages

Python SQL JavaScript C# R

Tools & Frameworks

LangChain/LangGraph Databricks MLflow DevOps GitHub Copilot Cursor Docker Git PostgreSQL Pandas NumPy PySpark PyTorch TensorFlow LightGBM Scikit-learn HuggingFace OpenAI API Azure Flask FastAPI

Senior AI Engineer

BASF

Toronto, ON, Canada | Oct 2024 - Present

Lead Engineer for FOX Talk2Data, BASF's flagship AI data assistant used by 4000+ users. Winner of BASF's 2025 Digital Award.

Technical Leadership & Mentorship Fullstack & API Development Evaluation & Testing Stakeholder Communication Python

Enterprise AI Assistant: Technical lead for design and development of FOX, BASF’s AI assistant serving 4000+ users. Winner of BASF’s 2025 Digital Award.
Natural Language to SQL: Designed and developed a LangGraph agent that translates natural-language questions into Databricks SQL queries and returns tables and charts, runs deterministic analysis of the data, and answers from unstructured reports using RAG. New data and RAG sources onboard through configuration alone.
Grounded Executive Briefing: Shipped a monthly AI briefing for finance leadership (live since Sep 2026): deterministic SQL finds which KPIs moved and why, the LLM writes only the narrative, and validation rejects unsupported figures. Built on a shared analysis engine that also powers the chat assistant, so both always report the same figures.
Dashboard-Aware Chat: Connected the assistant to the Power BI report a user has open, passing the page's metadata from a React frontend through a FastAPI backend to the agent, so answers reflect the data on screen.
Evaluation: Built an evaluation framework with automated tests that track SQL correctness, tool selection, and latency for each assistant, catching regressions before release.
Agentic Coding System: Built an agentic coding assistant on top of GitHub Copilot and Cursor that autonomously implements features, runs tests, creates pull requests, and reviews code across production repositories. Architected with specialized sub-agents, instructions, skills, and custom prompts, enabling a single engineer to operate with the throughput of a small team.
AI Delivery Leadership: Co-owner of the Forward Deployed Engineering and Delivery Acceleration work package in BASF, which aims to accelerate the delivery of AI-powered features to production.

Generative AI Engineer

YT Copilot

Toronto, ON, Canada | Sep 2023 - Oct 2024

Developed a Chrome extension using Generative AI to enhance YouTube experience.

Fullstack & API Development Evaluation & Testing Python JavaScript MLflow

YouTube LLM Agent: Developed a Chrome extension that skips promotional segments, summarizes videos, and enables semantic search over YouTube content using an agentic LLM.
Context Window Management: Used a MapReduce-style chunking and aggregation pipeline to efficiently handle LLM context limitations for long YouTube videos.
Fullstack Development: Built a JavaScript-based frontend integrated with YouTube’s UI and deployed a Flask / PostgreSQL backend on Azure Kubernetes Service (AKS) with Docker, plus Ory Kratos authentication and Stripe billing.
Finetuning & Optimization: Improved response quality and reliability using tool calling, structured outputs, fine-tuning, and prompt engineering techniques such as chain of thought prompting.
CI/CD Pipeline: Implemented a CI/CD pipeline with Azure CLI for automated testing, deployment, and monitoring.

Machine Learning Engineer

Super GeoAI Technology Inc.

Toronto, ON, Canada | Oct 2021 - Jul 2023

Led team of 7+ engineers developing computer vision models for agriculture.

Technical Leadership & Mentorship Fullstack & API Development Evaluation & Testing Stakeholder Communication Python

Computer Vision: Led 7+ engineers to develop object detection models using PyTorch/TensorFlow for seed grading and crop yield estimation from phone/drone images, improving efficiency for agricultural partners by over 80%.
Model Enhancement: Applied transfer learning (COCO and Pascal VOC) and performed hyperparameter optimization (Ray Tune), improving accuracy by 10% for seed grading and 15% for wheat head detection.
System Design: Deployed using Flask integrated with a NextCloud-based storage platform and Ory Kratos authentication.
Leadership: Conducted technical interviews, led team meetings, and enforced high-quality implementation standards through structured code reviews.

Machine Learning Engineer

IBM

Toronto, ON, Canada | Jul 2021 - Jan 2022

Used NLP to identify conflicting or duplicated entries in SRS documents.

Python Git Pandas NumPy PyTorch

NLP: Used BERT-based models to perform NLP on Software Requirements Specification (SRS) documents to detect conflicting and duplicate requirements.
Model Optimization: Achieved a 20% increase in F1-score through fine-tuning and hyperparameter optimization of a Microsoft pre-trained model.

Instructor - Practical Deep Learning (CIND850)

Ryerson University

Toronto, ON, Canada | Sep 2021 - Dec 2021

Mentored 20+ graduate students on applied deep learning projects.

Technical Leadership & Mentorship Python Pandas NumPy PyTorch

Mentorship: Mentored 20+ graduate students on applied deep learning projects, reviewing model architectures and ensuring best practices.
Assessment: Graded assignments and exams and provided targeted feedback on deep learning implementations and results.
Curriculum Development: Designed lab exercises/solutions, emphasizing practical applications of deep learning.
Lecture Delivery: Delivered lectures and tutorials on core deep learning concepts, ensuring clarity and practical relevance.

Quantitative Data Scientist

Electronica AI

Toronto, ON, Canada | Jun 2014 - Aug 2019

Designed and implemented high-frequency trading algorithms with daily order flow of $10M+.

Fullstack & API Development Evaluation & Testing Stakeholder Communication Python C#

Trading Algorithm Development: Collaborated with 5+ engineers to design and implement high-frequency trading algorithms with a daily order flow of $10M+, leveraging machine learning to improve predictive accuracy.

Quantitative Data Analyst

Alkemi Network

Toronto, ON, Canada | Aug 2016 - Oct 2018

Validated automated financial trading strategies and conducted stress testing on exchanges.

Evaluation & Testing Python C# Git

Validation and Stress Testing: Validated automated financial trading strategies based on intra- and inter-exchange price discrepancies and conducted stress testing on exchanges to identify bottlenecks and vulnerabilities in live-like scenarios.

Wikipedia Article Question-Answering System

Aug 2024 - Sep 2024

NLP pipeline over Wikipedia articles for NER and Question Answering using LangChain, HuggingFace Transformers, and spaCy.

Python Docker Git HuggingFace OpenAI API

Natural Language Processing: Developed an NLP pipeline over Wikipedia articles for Named Entity Recognition (NER) and Question Answering (QA) using LangChain, HuggingFace Transformers, and spaCy.
Similarity Search & Embedding Generation: Implemented FAISS-based similarity search and embeddings to retrieve relevant text chunks efficiently from large corpora.
System Deployment: Deployed the application using Flask and Docker, with unit tests to ensure reliability.

Natural Language SQL Querying System

Jul 2024 - Aug 2024

Natural language interface that converts user queries into SQL commands using LangChain and HuggingFace Transformers.

Python SQL Git PostgreSQL OpenAI API

SQL Generation: Built an NLP solution that converts user queries into SQL commands using LangChain and HuggingFace Transformers.
API Development: Implemented RESTful APIs in Flask. Used SQLAlchemy with PostgreSQL for schema management, migrations, and query execution.

Alzheimer's Disease Detection

Jul 2021 - Oct 2021

Deep learning model for Alzheimer's disease detection from MRI scans using TensorFlow and Keras.

Evaluation & Testing Python Git Pandas NumPy

Deep Learning Image Classification: Developed a deep learning model for Alzheimer’s disease detection from MRI scans using TensorFlow and Keras. Compared architectures such as DenseNet121, ResNet50, and VGG16.
Data Augmentation: Leveraged datasets including OASIS and ADNI, with preprocessing and augmentation to improve generalization.
Model Optimization: Applied transfer learning and Bayesian hyperparameter optimization to achieve a notable 10% accuracy boost in Alzheimer’s disease detection.

Coleridge Initiative - Data Citation Extraction

Mar 2021 - Jun 2021

BERT-based text-to-text transformer model with CRF for extracting data citations. Kaggle competition.

Evaluation & Testing Python Git Pandas NumPy

Natural Language Processing: Developed a BERT-based text-to-text transformer model in PyTorch with CRF for extracting data citations.
Data Preprocessing: Used NLP techniques with spaCy and Gensim for tokenization, stemming, and stop-word removal.
Competition Achievement: Achieved a top 5% rank, earning a silver medal in the competition.

Live Demo

Riiid! Answer Correctness Prediction

Oct 2020 - Jan 2021

Transformer and LightGBM models to predict student answer correctness based on historical performance. Kaggle competition.

Evaluation & Testing Python Git Pandas NumPy

Time-Series: Compared Transformer and LightGBM models to predict whether a student will answer a question correctly based on their historical performance.
Model Efficacy: Attained an AUC score of 80%, demonstrating strong predictive performance.
Competition Achievement: Achieved a top 3% position in the Kaggle competition, earning a silver medal.

Live Demo

Walmart Sales Forecasting

May 2020 - Aug 2020

LSTM and LightGBM models to forecast 28-day sales for 3,049 Walmart products. Master's thesis project.

Evaluation & Testing Python Git Pandas NumPy

Time-Series: Developed LSTM and LightGBM models to forecast 28-day sales for 3,049 Walmart products using historical and explanatory features.
Feature Engineering: Engineered additional features to capture seasonal patterns and store-level effects, improving forecast quality.
Data Analysis: Analyzed M5 competition data spanning ten Walmart stores across CA, TX, and WI from 2011 to 2016.
Optimization: Applied Bayesian optimization to fine-tune model hyperparameters.

Repository

Traffic Sign Recognition

Mar 2020 - Apr 2020

CNN to classify German traffic signs achieving 98.14% accuracy on Google Maps Street View images.

Evaluation & Testing Python Git Pandas NumPy

Image Classification: Designed and trained a Convolutional Neural Network (CNN) to classify German traffic signs, achieving 98.14% accuracy on Google Maps Street View images.
Data Augmentation: Implemented data augmentation techniques such as zoom, shear, and rotation to improve robustness and generalization.
Comparative Analysis: Compared CNN performance against ANN, Random Forest (RF), and Support Vector Machine (SVM) baselines.
EDA & Model Evaluation: Performed exploratory data analysis and evaluated models using precision-recall curves, heatmaps, and loss curves.

Repository

Genetic Variation of miRNA Primary Transcripts

Apr 2017 - Aug 2017

Investigation of genetic differences in miRNA primary transcripts across human populations. Undergraduate thesis project.

R Git

Genetic Study: Investigated genetic differences in miRNA primary transcripts across human populations, analyzing 1,127 SNPs from 1,881 miRNA transcripts to identify key trends.
Data Analysis: Used variance and Fixation Index (FST) to quantify genetic differentiation, comparing “average of ratios” and “ratio of averages” formulations for genome-wide FST.
Genetic Research: Showed that African populations exhibit the highest degree of genetic variation, supported by dendrogram-based analyses.

Repository

Master of Science

Ryerson University

Data Science and Analytics | Toronto, ON, Canada | Sep 2018 - Aug 2020

Supervisor: Professor Konstantinos Georgiou
Thesis: Monthly Sales Forecasting for Walmart with Advanced Machine Learning Techniques

Honours Bachelor of Science

University of Toronto

Bioinformatics (Specialist), Computer Science (Major) | Toronto, ON, Canada | Sep 2013 - Jun 2017

Graduated with high distinction; University of Toronto Scholar Award

Supervisor: Professor Zhaolei Zhang
Thesis: Detection of Genetic Variations in miRNA Primary Transcripts Using Statistical Analysis