Senior Applied AI Scientist · NLP · LLM Systems
Fatemeh Rahimi
I design and ship applied AI systems for document understanding, retrieval, evaluation, and workflow automation. Over 7+ years, I've worked across biomedical NLP, production ML services, and LLM-based agent workflows, with a focus on building systems that are useful, reliable, and grounded in real-world constraints.
Featured Projects
Production and research highlights
Multi-agent Support Automation
Pythonic AI · 2025 – present
Architected an LLM-powered multi-agent support workflow to triage incoming requests, execute routine code and configuration changes, and escalate ambiguous cases to humans. Combined tool-using agents, retrieval, and human-in-the-loop review to reduce manual engineering effort and speed up support resolution.
Production NLP Services for Title and Escrow Workflows
Pythonic AI · 2022 – present
Led development and deployment of 3 production NLP services for document-heavy title and escrow workflows, supporting extraction and classification across insurance, loan, mortgage, and signed closing documents. Served as primary technical owner for one service and built evaluation pipelines to measure model quality across releases and support reliable iteration.
Biomedical NLP Research
Imagia · 2020 – 2022
Researched and built NLP pipelines for clinical report analysis — including de-identification, named entity recognition, and measurement extraction. Applied transformer-based models (BioBERT, ClinicalBERT) to real-world hospital data.
MTLV: Multi-task Learning Library
Dalhousie University · 2020 – 2021
CLI library for exploring four multi-task learning architectures with pre-trained language models. Proposed a novel task-clustering approach that improved cross-task transfer. Published at ACM DocEng-2021; served as M.Sc. thesis.
Open Source
Contributions to public projects
Added two new OneDrive tools to this MCP server for Microsoft 365 — move/rename files and folders in a single PATCH request, and create new folders with conflict behavior support. Closes a workflow gap where users couldn't reorganize their OneDrive without leaving the AI assistant.
Added four new tools to this MCP server for Microsoft 365 — create, create child, rename, and delete mail folders via the Microsoft Graph API. Closes a workflow gap where users couldn't set up destination folders before moving messages.
Work Experience
7+ years in NLP, AI, and production ML
Senior NLP Scientist
Feb 2022 – PresentPythonic AI · Milwaukee, WI, US (Remote)
- –Led development and deployment of 3 production NLP services for title and escrow workflows, enabling extraction and classification across insurance, loan, mortgage, and signed closing documents; served as primary technical owner for one
- –Architected an LLM-powered multi-agent workflow with tool orchestration and human-in-the-loop escalation to automate support email triage and resolution
- –Built evaluation pipelines for LLM agents and production NLP systems, measuring tool-calling correctness and model quality to support more reliable iteration across releases
- –Developed automation tooling to streamline internal workflows across data labeling, sprint planning, and QA
- –Coordinated cross-functional annotation workflows and mentored team members on NLP and applied AI best practices
Applied Research Scientist – NLP
Nov 2020 – Jan 2022Imagia · Montreal, Canada (Remote)
- –Built NLP pipelines for hospital clinical report analysis — de-identification, named entity recognition, and measurement extraction
- –Applied and evaluated transformer-based models (BioBERT, ClinicalBERT) on real-world biomedical datasets
- –Designed production-grade CLI tools for end-to-end model inference on clinical documents
NLP Research Assistant
Mar 2019 – Apr 2021Dalhousie University · Halifax, Canada
- –Applied and evaluated deep language models (BERT, BioBERT, BlueBERT) on downstream NLP tasks with rigorous experimental design
- –Developed MTLV — a multi-task learning library with 4 architectures and a novel task-clustering method (published at ACM DocEng-2021)
Data Analyst
Aug 2018 – Feb 2019Ayten Company · Shiraz, Iran
- –Worked on social network analysis using Python, Neo4j, and MongoDB
- –Practised Scrum within an R&D software team
Skills
Core technologies and competencies
Languages
LLMs & GenAI
ML / DL
NLP
Data & Experimentation
Infrastructure
Publications
Peer-reviewed research
MTLV: A Library for Building Multi-task Learning Architectures for NLP
Rahimi, F., Milios, E., Matwin, S.
DocEng 2021 — ACM Symposium on Document Engineering · 2021
Talks
Public talks and presentations
March 2025
Compound AI Systems
Explores the shift from thinking about AI as isolated models to understanding them as complex systems. Covers how system-level thinking enables smaller, more efficient AI that outperforms simplistic large-scale models — with a focus on safety, optimization, and real-world performance.
January 2025
Trends in LLM Collaboration — Debaters and Judges in Agentic AI Systems
Covers the latest trends in LLM collaboration, highlighting agentic AI systems. Explores how collaboration, debates, and LLM-as-a-judge techniques improve AI workflows, with practical strategies for real-world implementation.
Education
Academic background
M.Sc. Computer Science
Dalhousie University · Halifax, Canada
GPA 4.07/4.3 · Thesis: MTLV multi-task learning library
B.Sc. Software Engineering
Shiraz University · Shiraz, Iran
Senior project: Kitchen Safety IoT system