Software & Machine Learning · Montréal

Building quiet systems
for noisy problems.

Portrait of Aref Afzali

I build, ship, and research ML-driven products — from co-founding Qliqless.ai to retrieval systems in production to graph research at Concordia.

Co-Founder · ML Engineer · Researcher

Montréal · QC M.Sc. Concordia Est. 2017
Roles
4+
Publications
02
Book
01
Years shipping
4+

§ About

On the work.

Aref is a software and ML engineer, co-founder, and graduate researcher. Most recently he co-founded Qliqless.ai, where he shipped a multi-tenant LLM conversational agent and led the infra end-to-end. Before that he built retrieval and recommendation systems at XRaise.ai, led an airline-ticketing platform from design to production as team lead at RomaParvaz, and solved vehicle-routing and geocoding problems at Carriot.

At Concordia he is completing an M.Sc. under Prof. Hovhannes Harutyunyan on broadcasting algorithms over k-path graphs. He has published on spiking neural networks in medicine and co-authored an open-source discrete-mathematics textbook. He’s happiest when research depth and production taste meet in the same project.

§ Trajectory

Where the work has landed.

  1. Apr 2025 — Oct 2025 Co-Founder & ML Engineer Qliqless.ai Co-founded and shipped an LLM conversational agent MVP; led architecture and infra end-to-end.
    • Built and shipped an AI conversational agent MVP that maps natural-language requests to real business actions via tool endpoints.
    • Designed a multi-tenant architecture using MCP to provision per-business tool endpoints and customer-specific agent configurations.
    • Containerized and deployed services with Docker/Kubernetes on AWS (EC2) behind an NGINX reverse proxy for reproducible environments and scalable rollout.
    • Delivered multiple end-to-end demos to pilot users and iterated weekly on product feedback.
  2. Dec 2024 — Feb 2025 R&D Engineer XRaise.ai Built a RAG-based company matcher and LLM pipelines for VC & grant workflows.
    • Built a RAG recommender with LlamaIndex + Qdrant matching startups to similar YC/TechStars companies; integrated Exa Search API for live company data.
    • Implemented a hybrid retrieval router combining vector similarity and keyword retrieval to improve coverage across sparse and exact-match queries.
    • Evaluated retrieval quality with DeepEval and RAGAS to tune top-k and compare strategies.
    • Architected an LLM-assisted pipeline extracting structured fields from VC and grant forms using HTML analysis, multimodal agents (Browser-Use), and HyperBrowser.
  3. Mar 2022 — Mar 2023 Software Team Lead RomaParvaz Travel Agency Led end-to-end delivery of a production airline ticketing platform.
    • Led end-to-end delivery of an airline ticketing platform from design to production using Scrum; coordinated backend, frontend, and integrations.
    • Architected a microservices, event-driven backend (NestJS) with React frontend and PostgreSQL/MongoDB/Redis.
    • Integrated multiple GDS providers (Amadeus, TravelPort, Gabriel) with resilient handling for third-party latency and failures.
    • Owned production delivery: Docker + Compose, GitLab CI, NGINX with TLS + rate limiting, Cloudflare DNS/CDN, centralized logging with the ELK stack.
  4. Jul 2020 — Apr 2021 Data Specialist Carriot Built routing optimization and geospatial analytics for a logistics platform.
    • Built a VRP optimization API in Flask using Google OR-Tools plus custom heuristic/metaheuristic methods, supporting PDPTW, time windows, and multi-depot routing.
    • Developed a Persian address geocoding service using Sent2Vec embeddings and clustering to improve logistics address resolution.
    • Implemented vehicle behavior analytics (stop-type detection, acceleration-axis calibration) on speed, acceleration, and geospatial signals.
    • Built geospatial heatmaps to communicate demand patterns and fleet movement to stakeholders.

§ Projects & Research

A selection, plainly told.

research · 2024

MGCLTransformer — Multimodal Graph Recommendation

Graph Transformer + Multimodal Graph Contrastive Learning on user–item interactions with images and titles. +21.4% Hit Ratio, +44.1% NDCG over baselines.

  • PyTorch, PyTorch Geometric, SentenceTransformer, ResNet-50
  • Multimodal features fused with contrastive objectives

Plate II · Thesis

thesis · 2026

Broadcasting Algorithms on K-Path Graphs

M.Sc. thesis at Concordia with Prof. Hovhannes Harutyunyan — studying broadcast time and algorithms on the k-path family of graphs.

  • Graph theory, algorithm design, combinatorial optimization
  • Focus on structural properties governing broadcast schemes

product · 2021

VRP Optimization API

Flask API solving Vehicle Routing Problems at Carriot — PDPTW, time windows, and multi-depot cases via OR-Tools with custom heuristics.

  • Google OR-Tools + heuristic/metaheuristic search
  • Feasibility and objective tuning for real logistics

product · 2022

Basket Recommender — B.Sc. Thesis

Association-rules recommender with Flask-RESTful backend, VueJS frontend, PostgreSQL, and Power BI analytics. Selected as 2nd-place faculty B.Sc. thesis.

  • Market-basket analysis and targeted customer recommendations
  • End-to-end: data → model → API → UI → BI dashboard

§ Publications

What’s in print.

  1. Afzali A., Bashizade M., Akbarein H. (2023).

    The Application of Spiking Neural Network in Schizophrenia.

    1st International Congress of Artificial Intelligence in Medical Sciences (AIMS 2023) — Poster

  2. Shayegh B., Afzali A., MohammadHashemi S., MohammadTaheri K., Mohammadi S. (2021).

    Discrete Mathematics: An Introduction with an Academic Approach.

    Open-source book manuscript

§ Notes

Recent writing.

No published notes yet. Coming soon →

View all notes →

§ Teaching

On passing it along.

Aref has taught and mentored across computer-science disciplines — from introductory programming to applied statistics — and co-authored an open-source discrete-mathematics textbook (2021) used by undergraduates as an approachable on-ramp to proofs and combinatorics.

§ Off the clock

A life in two registers.

Outside engineering, Aref is a 4th-dan black belt and instructor in Kyokushin Karate — a long-running counterweight to the screen-bound half of his life. He plays guitar, reads broadly, and is fluent in Persian and English, with elementary French and German.

§ Contact

Say hello.

Or find me on