Post Job Free
Sign in

Generative AI & Agentic Systems Engineer

Location:
Toronto, ON, Canada
Posted:
November 19, 2026

Contact this candidate

Resume:

Mohamed Hafez

AI Engineer Generative AI, Agentic AI & LLM Evaluation Toronto, ON, Canada

+1-647-***-**** *******.*.********@*****.*** mohamednhafez.com linkedin.com/in/mohamed-nader555 github.com/Mohamed-Nader555 Professional Summary

AI Engineer with 4 years of applied AI/ML and NLP experience, now designing, building, and deploying production Generative AI and Agentic AI solutions as Founding AI Engineer at a Toronto healthcare-communication startup. Builds multi-agent systems, conversational AI, AI copilots, tool/function calling, prompt engineering and workflow orchestration, and Retrieval-Augmented Generation (RAG) services, with AI evaluation frameworks, observability, reliability, and Responsible AI (privacy, PII protection, governance) built in from the start. Master’s research on LLM and transformer evaluation for PII-aware NER, with a peer-reviewed paper accepted at CASCON 2026. Prior production ML (fraud detection, forecasting) and enterprise backend integration for regulated financial-services clients. Mentor, teaching assistant, and clear communicator to technical and business stakeholders.

Work Experience

Founding AI Engineer Sep 2026 – Present

Eklan – AI communication-training platform for healthcare professionals Toronto, ON

• Own the platform’s Generative AI features end to end on a small founding team: model integration, TypeScript/Node.js/Next.js REST APIs, MongoDB (NoSQL) data models, React UI, evaluation, and feature-flagged CI/CD release from staging to production; translate founder requirements into architecture and API contracts the mobile team builds on.

• Architected and deployed a real-time conversational AI system on the Gemini Live API: browser audio streaming secured by short-lived, config-locked tokens, latency-focused performance optimization of voice-activity detection, interruption handling, and function calling that drives scenario phases, with app-side orchestration so the model cannot skip steps.

• Built a multi-agent system in which each AI agent runs as its own live session and a low-latency router (director) model sends every user turn to the right agent, with speculative cueing, safe fallback, and guardrails that stop off-script or wrong-speaker output; built and load-tested for scalability across concurrent sessions.

• Designed a multi-stage LLM evaluation pipeline (transcription, pronunciation scoring, LLM language and communication scoring, rule-based fluency, LLM coaching) that runs during the conversation, with retries, locks, model fallback on rate limits, per-stage status and error reasons, and a re-grading consistency check for reliability.

• Shipped an on-demand AI copilot for personalized coaching using structured outputs against a versioned schema, versioned prompts, a rule-based output checker, English/Korean PII redaction before anything is stored or sent on, an eval suite against the live model, and a leased, retrying, event-driven background job queue (agentic workflow) that turns a resolved chat into personalized practice.

• Replaced a single generation prompt with composed, per-task prompts (user profile, scenario template, LLM-selected focus, task rules) under a token budget; built LLM pipelines over unstructured data that extract each learner’s mistakes from class transcripts and generate PowerPoint and Word documents.

• Built an automated test harness for live voice agents (synthetic user audio, latency, turn-taking and overlap checks, plus an LLM-as-judge) so every prompt or engine change was measured before release; production monitoring and observability with Sentry, Vercel logs, and PostHog dashboards; spec-driven agentic development with Claude Code and MCP. Graduate Researcher – Machine Learning & NLP (CERAS Lab) Sep 2024 – Apr 2026 York University Toronto, ON

• Built ASC-PIE, a PII-NER corpus and evaluation framework: 333,109 examples and 2,025,878 entity mentions unified from five public datasets plus a synthetic component under one 19-type schema, with data-validation and label-quality checks (Python, PyTorch, Hugging Face Transformers/Datasets).

• Benchmarked encoder (BERT, RoBERTa, ModernBERT), encoder–decoder (FLAN-T5), and decoder-only LLMs (Llama 3.1 8B, Qwen 2.5 7B) under one protocol, comparing fine-tuning with in-context learning on strict span-level precision, recall, F1, latency, output validity, and robustness; best strict F1 99.1%.

• Developed SPRINT-PP, a privacy-safe continual-learning method (paper accepted at CASCON 2026) reaching 83.5% accuracy without storing raw historical PII, ahead of replay (76.5%) and a naive baseline (14.6%).

• Kept every experiment reproducible and auditable (schemas, prompts, configs, prediction traces); also a Teaching Assistant in Systems Architecture and Data Visualization, leading labs and mentoring students. Software Engineer – ECM Consultant Aug 2023 – Sep 2024 BASS (financial-services clients) Cairo, Egypt

• Implemented OpenText Documentum workflows for document intake, routing, review, and retention for banking clients in regulated, PII-sensitive environments; wrote SQL/DQL validation and metadata-quality checks for compliance and governance.

• Built Java and Python backend services and internal REST APIs with standardized JSON schemas and authentication for enterprise system integration; ran log-based root-cause analysis, wrote runbooks and documentation, and mentored junior engineers. AI & ML Specialist Jun 2021 – Aug 2023

DotPy Cairo, Egypt

• Delivered 10+ applied ML projects (fraud detection, forecasting, segmentation, recommendation, sentiment analysis, image classification) with Python, SQL, scikit-learn, TensorFlow, and PyTorch, owning the AI/ML lifecycle from scoping and feature engineering to evaluation, model deployment, and handover.

• Shipped API-based model services and Dockerized prototypes with CI/CD, versioning, and code review; defined serving contracts (authentication, rate limits, error handling) with engineering teams; early LLM application prototyping.

• Designed and taught Python and AI/ML workshops and mentored junior developers, explaining technical concepts to mixed technical and non-technical audiences.

Freelance Android Developer Jul 2019 – Jun 2024

Self-Employed / Client Projects Remote

• Built 9 Android applications in Kotlin/Java (4 published through client-owned Google Play listings), including apps with OCR, computer vision, and API-based ML inference.

Selected Projects

Northstar – Grounded RAG Service Python, FastAPI, LangChain, Chroma vector database, sentence-transformers, Docker, pytest

• Built a RAG API over enterprise-style policy documents: PDF/Markdown ingestion, metadata-rich chunking, local embeddings in a vector database, and distance-filtered retrieval, so no document text leaves the machine to be embedded.

• Every answer cites document, page, and chunk, and the service refuses before generation when evidence is weak; configurable LLM backend, pytest coverage of happy, refusal, and malformed paths, one-command Docker Compose deployment. Skills

Generative & Agentic AI: LLM applications, Agentic AI, multi-agent systems, agent orchestration, tool/function calling, prompt engineering, structured outputs, RAG, AI copilots, conversational AI, LangChain, Hugging Face Transformers, vector databases (Chroma), OpenAI and Gemini APIs, guardrails, MCP

Evaluation & Responsible AI: AI application evaluation frameworks, eval suites, LLM-as-judge, grader calibration, precision/recall/F1, latency and load testing, PII detection and redaction, privacy by design, governance, auditability Languages: Python, TypeScript, JavaScript, Java, SQL, Kotlin Engineering: Node.js, Next.js, React, FastAPI, REST APIs, WebSockets, event-driven background jobs, role-based access control, application security, Docker, CI/CD, Git/GitHub, feature flags, testing

Data & Platforms: MongoDB Atlas (NoSQL), SQL Server, structured, semi-structured and unstructured data, Vercel (serverless, cron), Sentry, PostHog; AWS and GCP (working knowledge); Spark (basic) ML: PyTorch, TensorFlow, scikit-learn, pandas, NumPy, fine-tuning, in-context learning, NER, continual learning, computer vision Education

York University Sep 2024 – Apr 2026

M.A. Information Systems & Technology (MAIST), degree awarded 2026 Toronto, ON

• Thesis: ASC-PIE: An Evaluation Framework for PII-Aware Named-Entity Recognition (YorkSpace, 2026). Studies how PII extraction stays accurate and maintainable as PII types expand: a unified corpus with synthetic coverage of rare PII without real personal data; encoder, encoder–decoder, and decoder-only models under supervised fine-tuning and in-context prompting; and evaluation of extraction quality, robustness, output validity, efficiency, and knowledge retention.

• Publication: SPRINT-PP, privacy-safe continual learning for PII extraction without storing raw historical examples. Accepted at CASCON 2026. Helwan University Sep 2018 – Jun 2022

B.Sc. Computer Science & Artificial Intelligence Cairo, Egypt

• Excellent with Honors, ranked 2nd in class, GPA 3.72/4.0; ICPC contestant 2019, 2020, 2021.

• Graduation project (A+) – Mind’s Eye: assistive smart-glasses system for visually impaired people and people living with Alzheimer’s that turns the scene into speech: face detection and recognition with spoken memory cues for familiar people, object and label detection, scene description, OCR, and text-to-speech (Android/Java, Arduino, OpenCV, Tesseract OCR, cloud vision over REST).

• Delivered 80%+ of the end-to-end system (Android app, cloud integrations, Arduino device-to-app communication layer, interface contracts and error handling) and built Egyptian currency recognition 100% from scratch, with OCR/TTS confirmation, because no existing solution recognized Egyptian banknotes.

Certifications & Languages

AWS Machine Learning Foundations, Udacity (2022) English (professional), Arabic (native)



Contact this candidate