Software Engineering · Machine Learning · Systems Design

Hi, I'm Shilo — a Mechatronics Engineer building at the intersection of AI and Robotics.

Mechatronics Engineering student at the University of Waterloo, currently interning as an ML Research Engineer and Software Engineer. I build AI-powered products, contribute to autonomous drone systems, and compete in hackathons.

Shilo Jeyaraj

Experiences

Internships and team roles where I've shipped products, trained models, and built systems.

Coincidence Labs
INTERNSHIP · APR 2026 – PRESENT

Coincidence Labs

Machine Learning Engineering Intern

Cohere Labs
INTERNSHIP · JAN 2026 – PRESENT

Cohere Labs

Machine Learning Research Engineer

Perceivable Design Studios Inc.
INTERNSHIP · JAN 2026 – PRESENT

Perceivable Design Studios Inc.

Machine Learning Research & Development Intern

Friedmann AI
FULL-TIME · SEP 2025 – PRESENT

Friedmann AI

Software Engineering Intern

Normative
INTERNSHIP · JAN 2025 – APR 2025

Normative

Software Engineer Intern

Waterloo Aerial Robotics Group (WARG)
EXTRACURRICULAR · MAY 2025 – PRESENT

Waterloo Aerial Robotics Group (WARG)

Autonomy Team Member

Projects

Explorations at the intersection of ML, product design, and full-stack.

Proof

Proof

ReactTypeScriptFastAPI+3
Dermalens

Dermalens

Next.jsFastAPIGemini API+4
Coursely

Coursely

Next.jsOpenAILangChain+3
Brain Battle

Brain Battle

ReactNext.jsNode.js+3
Mid Chats

Mid Chats

Next.jsFastAPIWebSockets+3
Solar Panel Calculator

Solar Panel Calculator

PythonFlaskREST API+2
Gym Posture Corrector

Gym Posture Corrector

PythonOpenCVMediaPipe+2

OCR Document Analyzer

PythonOpenAI APIOCR+2

Nutrition/Fitness Plan Builder

PythonFlaskGemini API+2

Research

Exploring the frontiers of machine learning, LLM adaptation, and intelligent systems.

Memory Retrieval Strategy Matters: A Comparative Study of Episodic Memory Backends for Reflexion-Style LLM Agents

Machine Learning
In Review
May 2026

Memory Retrieval Strategy Matters: A Comparative Study of Episodic Memory Backends for Reflexion-Style LLM Agents

We show that retrieval ordering within a backend class accounts for a larger performance gap (12pp) than the choice of backend itself. Vector DB leads on first-attempt success; SQL converges once failure-first ordering is applied.

Read paper →