Founder
& AI engineer

Toronto, Canada
43.6604° N, 79.3893° W

Frontier AI, human by design.

I’m building the next generation of human-centric, AI-powered companies and designing the frontier AI frameworks they run on. AI engineer at AXL, a Toronto venture studio. Computer Engineering at Waterloo.

Haoran Zhu

01Profile

I’m a Computer Engineering student at the University of Waterloo. I’ve shipped features for Baidu’s ERNIE Bot, and I led Oliver, an AI tutor that remembers the conversation. The paper on it appeared at IEEE ICEIT 2025.

02Now

Agents do the legwork. A person makes the call.

This is how I like to build agent systems. A planner breaks the task down, other agents retrieve, act and check each other’s work, and anything that matters stops at a person before it ships. The run shown here is a simulation, and this time the person is you.

  • AXL

    Building human-centric, AI-powered companies at a Toronto venture studio.

  • McGill × CEIMIA

    Measuring how reliably LLMs retrieve and generate for climate and disaster analysis.

agent run / human in the loopsimulated

    03Selected work

    Selected work

    Each of these four projects runs live in your browser. Try them.

    1. 01 / 04 LLM application

      Oliver

      A virtual teaching assistant that remembers.

      A course Q&A and management system built on large language models. Customizable prompts and conversational memory let it follow a student’s train of thought instead of answering every question cold. I designed, built and evaluated it as lead researcher, and the paper was published at IEEE ICEIT 2025.

      • Python
      • PyTorch
      • FastAPI
      • ChromaDB
      • RAG

      oliver / sessionscripted replay

      memory

    2. 02 / 04 Data structures · C++

      LLMs‑Tokenization

      Turns words into numbers and sorts out the collisions.

      Before a language model reads anything, text has to become token IDs. This is that first step, written from scratch in C++: a hash table that maps each word to a slot and resolves collisions with separate chaining.

      • C++
      • Hash table
      • Separate chaining
      • Make

      hash table / m = 7type to tokenize

      1. 03 / 04 Algorithms · C++

        Navigation Graph

        Finds the fastest route across town.

        A weighted, undirected graph of city intersections where each edge is a travel time. Dijkstra’s algorithm on a heap-backed priority queue finds the lowest-cost route between any two points.

        • C++
        • Dijkstra
        • Priority queue
        • Heap

        dijkstra / shortest pathhover or use arrow keys · click or Enter moves the starttap an intersection to route there

      2. 04 / 04 Spatial indexing

        Quadtree Range Search

        Find what’s nearby without looking everywhere.

        A spatial index that keeps splitting the plane into four, so a range search only opens the cells it overlaps. The same structure draws the background at the top of this page.

        • Quadtree
        • Range search
        • Spatial index

        quadtree / range searchclick, drag or press Enter to add pointstap or drag sideways to add points

      04Research

      Enhancing contextual understanding in AI‑powered tutoring: evaluating the Oliver system for effective learning support.

      Venue
      ICEIT 2025, the 14th International Conference on Educational and Information Technology
      Publisher
      IEEE
      Authors
      Haoran Zhu, Michael Cooper‑Stachowsky, Zille Huma Kamal
      Role
      Lead researcher and developer
      1. A

        Studied how AI tutors lose the thread, and how memory management and dynamic responses can keep it.

      2. B

        Designed and built Oliver’s core architecture around prompt engineering and conversational memory.

      3. C

        Evaluated it through testing and analysis. The results showed better student engagement and outcomes.

      Read on IEEE Xplore

      05Experience

      Where I’ve worked

      1. Aug 2026 – Present AXL AI Engineering · Co-op
        • AXL is a Toronto venture studio building the next generation of human-centric, AI-powered companies, turning frontier AI research into real-world businesses.
        • Returned to AXL Labs, the engineering team behind every venture, to prototype and ship the AI products that take new companies from concept to market.
      2. Jan – May 2026 AXL AI Software Engineering · Co-op
        • Joined AXL Labs to help build the next generation of human-centric, AI-powered companies alongside founders, designers and researchers.
        • Built full-stack products for the studio’s ventures, taking them from first prototype to working systems.
        • Brought frontier AI into those products with LLMs and retrieval-augmented generation, and built internal tools and shared components used across the portfolio.

        Frontier AI · LLMs · RAG · Full-stack

      3. Oct 2025 – Present McGill University Research Assistant
        • Working with CEIMIA on a retrieval-augmented generation (RAG) evaluation framework for climate and disaster analysis. It measures how reliably large language models retrieve and generate answers in this domain.

        RAG · Pinecone · LLM evaluation

      4. May – Dec 2025 WAT.ai Technical Project Manager · Internship
        • WAT.ai is the student design team of the Waterloo Data and Artificial Intelligence Institute.
        • Ran AI projects as technical project manager and worked on joint research into ReAct agents, LangGraph and multi-agent systems.

        RAG · LangGraph · ReAct agents · Azure

      5. May – Aug 2025 HVE Healthcare Assessments Full-Stack Developer · Co-op
      6. Sep – Nov 2024 University of Waterloo AI in Education Research & Development Assistant
        • Led the development of Oliver, a RAG-based tutoring system, and studied how RAG affects conversational memory. The findings were published at ICEIT 2025.
        • Built LLM-based solutions for automated tutoring, grading and feedback generation.
        • Measured in Python how well language models tutor and grade.
        • Developed the front- and back-end interfaces that deploy these systems, then curated and analyzed the data from real user interactions.

        Python · PyTorch · FastAPI · SQLAlchemy · ChromaDB · RAG · Transformers

      7. Jan – Apr 2024 Baidu ERNIE Bot · Software Engineering, Full-Stack
        • Worked with the research team to gather feedback and iterate on the latest ERNIE Bot models.
        • Optimized code running on Azure VMs, and tuned applications for speed and scale.

        Python · PyTorch · REST · Azure · Docker · Django

      8. May – Aug 2023 Baidu ERNIE Bot · Machine Learning & Software Development Assistant
        • Helped maintain data sources and designed a data-management pipeline with Python, Docker and recurrent neural networks.
        • Applied deep-learning methods such as PP-LCNet and reinforcement learning to improve the learning process and data validation.
        • Shipped new ERNIE Bot features to Baidu’s apps through API deployment.

        Python · Docker · RNNs · API development

      9. May – Aug 2022 Yungang Law Firm Web Developer
      10. Nov 2019 – Present Halton Region Chinese Canadian Association Youth Leader · Part-time
        • Run competitions, events and orientations for the association.

        Communication · Innovation management

      06About

      The person behind it

      Portrait of Haoran Zhu
      fig. 1: quadtree decomposition

      I like the moment a research idea turns into something a person can use.

      I study Computer Engineering at the University of Waterloo. At Baidu I worked on ERNIE Bot, where I optimized models and built data pipelines. I also led Oliver, an AI tutoring system that uses conversational memory to keep track of context.

      I’ve worked in industry and in research labs, and in both places I cared most about whether a system stays fast under real load and whether people come back to it. Outside of code, I’ve been a youth leader with the Halton Region Chinese Canadian Association since 2019.

      1. 01

        AI application development

        LLM applications built for one job, like Oliver, with custom prompts, conversational memory and retrieval.

      2. 02

        Web development

        Front to back with Django, FastAPI, Jinja2, React and Vue.

      3. 03

        Database design

        Schemas, query optimization and performance tuning on PostgreSQL and MySQL: indexing, joins, window functions, stored procedures.

      Languages

      • Python
      • C/C++
      • Go
      • Java
      • JavaScript
      • HTML/CSS
      • MATLAB
      • Assembly
      • VHDL

      Frameworks & tools

      • PyTorch
      • RAG
      • ChromaDB
      • FastAPI
      • Django
      • React
      • Vue
      • Node
      • Docker
      • Azure DevOps
      • Linux
      • Git

      Download résumé More on GitHub