Mahdi Farzandway

Visiting Researcher

SPEAR Lab – Concordia University

Jan 2026 - Present

Collaborating with Prof. Tse-Hsun (Peter) Chen on LLM-based fault localization and automated program repair (Mitacs Globalink Research Award). Designing a multi-stage RAG-enhanced LLM pipeline for automated fault localization and program repair, evaluated on the Defects4J benchmark. Exploring chain-of-thought prompting strategies and retrieval index architectures integrating execution traces, logs, and commit history.

Lecturer

Razi University

Mar 2024 - Present

Teaching undergraduate-level courses in Artificial Intelligence, Data Science, and Machine Learning. Mentored students in applied machine learning and data science projects. Taught in Persian, class size: ~50 students.

NLP Researcher

System Verification Lab, University of Tehran

Nov 2020 - Present

Designed and developed NLP and LLM-based systems for automated fault localization and code repair. Solve complex problem-solving tasks by applying advanced techniques such as RAG, prompt engineering, PEFT, LoRA, and multi-step reasoning. Built and evaluated end-to-end machine learning pipelines, focusing on robustness, scalability, and real-world applicability. Worked with modern LLM frameworks and tooling (e.g., LLaMA, Gemma, BERT, Hugging Face, LangChain) to prototype and refine intelligent systems. Translated research outcomes into practical system designs with measurable performance improvements. Delivered validated results through peer-reviewed publication (AITest 2024) and an ongoing journal submission.

Lead Data Scientist

Pars Pack

Jun 2024 - Dec 2025

Led and scaled a cross-functional team of 5 data scientists and engineers, owning the end-to-end delivery of 4+ production AI and LLM-powered systems supporting core customer operations. Mentored junior data scientists and engineers through code reviews, pair work, and hands-on project guidance to improve delivery quality and technical growth. Owned the technical and strategic direction of an LLM- and RAG-based customer response platform built with LangGraph and FAISS vector database, processing 1.5M+ FAQ records with multi-agent orchestration for intelligent query routing and response generation. Delivered an 80% reduction in median customer ticket resolution time, directly improving customer satisfaction and operational efficiency at scale. Drove cost-efficiency as a first-class product metric, achieving a 35% reduction in inference costs through structured model evaluation, trade-off analysis, and iterative optimization. Processed 1.5M+ records via batch pipelines to extract high-frequency and unique queries for vector database enrichment and improved retrieval accuracy. Established a company-wide experimentation culture, defining success metrics, designing controlled rollouts, and ensuring decisions were backed by statistically reliable evidence. Improved system responsiveness by guiding architectural decisions around retrieval, embeddings, and query optimization. Built real-time monitoring and KPI dashboards covering accuracy, latency, cost, and customer outcomes, maintaining 99.5% production uptime. Acted as the primary data and AI partner to product, engineering, and customer support, translating business goals into measurable outcomes and technical roadmaps. Applied statistical inference techniques (hypothesis testing, confidence intervals) to evaluate model improvements and guide high-stakes product and deployment decisions. Presented performance results, risks, and strategic recommendations biweekly to the CEO, directly influencing product prioritization and investment decisions.

Data Scientist

Orpiva.ai

Feb 2023 - Jun 2024

Designed and developed an AI-powered system that ingests a product or business link and automatically generates short-form promotional videos. Built intelligent pipelines to extract and structure content from input links, enabling automated video generation in under 45 seconds. Developed AI-driven methods to generate engaging advertising copy tailored for social media platforms such as Instagram and TikTok. Created and maintained a library of 70+ reusable video templates adaptable to a wide range of products and links. Collaborated on and implemented the core AI and backend components of the system, ensuring scalability and production readiness. Delivered a data-driven solution that significantly streamlined digital marketing content creation.

Sr. Backend Developer

Wegobazzar

Feb 2017 - Jul 2018

Jr. Backend Developer

Tetra

Jun 2015 - Sep 2016