Skip to content
Ryan Chwiecko

Ryan Chwiecko

Software engineer working on backend systems and AI infrastructure

London, Ontario, Canada
Graduating April 2027 and open to new-grad software engineering roles in Canada and the US

I build backend systems and the infrastructure that makes AI useful outside a demo: data pipelines, retrieval systems, and services that have to keep working at scale. Most recently I spent 16 months as a software engineering intern at J.D. Power, building analytics tools and data pipelines over automotive datasets with more than 10 million records.

Experience

Figure 1. Work and community roles drawn as concurrent processes. Lines that overlap ran at the same time, and an arrow means the role is ongoing.

Work experience

  1. Software engineering intern

    J.D. Power, May 2025 to August 2026

    • Built and deployed an analytics interface that lets clients query datasets of more than 10 million records through BigQuery.
    • Built a data visualization pipeline with Azure Data Factory and PySpark, and ETL pipelines in AWS.
    • Worked on privacy-compliance infrastructure for encrypting and deleting user data.
    • Helped modernize internal systems in React and Spring Boot, including search, data management, and admin workflows.
    • Tracked down and fixed failures in the automated test and smoke-test pipelines.

    Java, Spring Boot, React, BigQuery, PySpark, Azure Data Factory, AWS

  2. Software engineering intern

    Headstarter, August 2024 to April 2025

    • Built a platform for benchmarking and fine-tuning more than 10 LLMs on Groq infrastructure.
    • Built a retrieval-augmented generation system for analyzing codebases, using Pinecone and Hugging Face embeddings.
    • Developed backend services in Python and Flask, running as containers on cloud infrastructure.
    • Implemented multimodal AI features for recommendation and assistant apps.

    Python, Flask, Groq, Pinecone, Hugging Face, Docker

  3. Software engineering intern

    Predictify Pro, September 2023 to September 2024

    • Built backend APIs in Flask and MongoDB for a housing-price prediction platform with more than 500 users.
    • Designed a Python API for storing and querying user data in MongoDB, with encryption for sensitive fields.
    • Co-led development of an LSTM model in TensorFlow for predicting housing prices.
    • Led the front-end architecture in React, TypeScript, and Tailwind CSS.

    Python, Flask, MongoDB, TensorFlow, React, TypeScript

  4. Software engineer

    Ribbon Business, April 2023 to August 2023

    • Built a web calculator in React that compares the benefits and pay of working as a contractor versus an employee.
    • Designed the layout as Figma wireframes and confirmed it with the project manager before building.

    React, Figma

Leadership and community

  1. Vice President of Externals

    Computer Science Undergraduate Society, Western University, September 2026 to present

    • Lead external relations, industry partnerships, and professional-development programming for computer science undergraduates.
    • Organize technical panels and speaker events that connect students with working engineers.
    • Manage a team of three directors.
  2. Member and event volunteer

    Google Developer Group London

    • Help plan local tech events and work on community projects.

Projects

MCP Orchestrator

Python, AWS ECS Fargate, Lambda, API Gateway, S3

A platform for provisioning and managing custom Model Context Protocol servers in the cloud, designed so people without infrastructure experience can deploy one and idle servers stop costing money.

  • Containerized deployment and lifecycle management for each server on AWS.
  • Per-server configuration, cloud-based execution, and automatic shutdown of idle resources.
  • Built around lowering idle infrastructure costs and keeping deployment simple for nontechnical users.
Figure 2. Requests arrive through API Gateway, Lambda functions provision and configure servers, and each MCP server runs as its own Fargate task that’s shut down when idle.

Resumit

Next.js, Supabase, Stripe, LLM APIs

A web app that tailors a résumé to a job description using only the user’s verified professional information, with workflows designed to keep the language model from inventing claims.

  • Authentication and subscription billing built on Supabase and Stripe.
  • Language-model generation grounded in the user’s verified work history.
  • Workflows that reduce unsupported or hallucinated résumé claims.
Figure 3. The Next.js app handles sign-in and stored profiles through Supabase, billing through Stripe, and tailoring through an LLM API.

Codebase RAG

Python, Pinecone, Hugging Face

A retrieval-augmented generation system for asking questions about a codebase, built during my Headstarter internship.

  • Embeds source code with Hugging Face models and indexes it in Pinecone.
  • Answers questions with the most relevant code retrieved as context for the language model.

Source on GitHub

Figure 4. Code is embedded with a Hugging Face model and stored in Pinecone; each question retrieves the closest chunks as context for the LLM.

More projects

  • LLM evaluation platform

    TypeScript, Groq

    Benchmarks and compares more than 10 language models side by side on Groq infrastructure. Built during my Headstarter internship.

  • Vestique

    Python, Streamlit, SambaNova, OpenCV

    An AI wardrobe assistant that recognizes clothing from photos, tracks what gets worn, and recommends outfits. Won Best Use of SambaNova and Best Use of Streamlit at HackUTD 2024.

  • Pose-estimation app, Consensus Hacks 2024

    Python, computer vision

    A computer vision and pose-estimation app built with a team. Won Best Use of Polkadot, Best Use of Robonomics, and second place in the Unique Network track.

Research

My thesis looks at how LLM-based vehicle diagnostics hold up at fleet scale. It extends LLM-OBD, a recent paper from Prof. Fang’s group (IEEE Communications Standards Magazine, 2026), from a single-vehicle edge–cloud prototype to deployment across a fleet.

Undergraduate honours thesis, Western University, supervised by Prof. Fang Fang, since September 2026

  • Finding where the original pipeline stops scaling: retrieval and query augmentation that slow down linearly as data grows, LLM serving that handles one request at a time, and per-driver generation that rules out caching.
  • Designing a replacement for each: approximate nearest-neighbour search with reranking, batched LLM serving that puts safety-critical alerts first, and a split between diagnosis and personalization so diagnoses can be cached.
  • Building a fleet-scale simulator and benchmark to compare throughput, tail latency, cost per diagnosis, and response quality against a reproduced baseline.

Research interests. Distributed systems, scalable AI infrastructure, machine-learning systems, edge computing, adaptive inference, and cloud computing.

Figure 5. The proposed design: retrieval by approximate nearest-neighbour (ANN) search with reranking, batched serving that prioritizes safety-critical alerts, and cached diagnoses kept separate from per-driver personalization.

Papers I keep rereading

Contact

Email is the best way to reach me. I’m looking for new-grad software engineering roles in backend systems and AI infrastructure, in Canada or the US, starting after I graduate in April 2027. I’m also always happy to talk about system design or anything I’ve built.

[email protected]