Looking for a job

Kaushal Subramani

@kaushalsubramani

AI engineer, ~2 yrs building LLM and agent systems at startups. RAG, evals, fine-tuning, Python. IEEE-published. CS grad.

Canada

Message

What I'm looking for

I'm looking for an AI engineering role where I can own my work end-to-end and grow with the company long term. Close to two years of experience building LLM and agent systems at startups, from fine-tuning and deployment to RAG pipelines and evaluations, and I want to keep building AI that reaches real users.

At Wilfrid Laurier University, I build and test quantum-classical machine learning pipelines for decentralized multi-drone navigation. I replaced a DQN’s classical value network with a parameterized quantum circuit and benchmarked it against the classical baseline.

At Kinara Systems, I deployed and fine-tuned Qwen 2.5 3B on edge hardware, improving telemetry and log classification by 10% over baseline. I also built a self-hosted inference service for on-device models and redesigned PDF and PPT knowledge ingestion to improve retrieval precision.

At Dawson Partners, I built a LangSmith evaluation system for LangGraph agents that raised reliability by 33% on time-sensitive financial calculations. My work also includes a ROS2 Nav2 path-planning project that achieved 2–7% shorter paths, 97% goal-reach, and zero collisions against matched NavFn baseline goals.

Experience

Work history, roles, and key accomplishments

Current

Quantum Machine Learning Research Assistant

Current

Wilfrid Laurier University

Jun 2026 - Present (3 months)

Built and tested quantum-classical ML pipelines for a MARL framework driving decentralized multi-drone navigation. Replaced the DQN's classical value network with a parameterized quantum circuit (PQC) as the Q-function approximator, benchmarking returns and trainable parameter counts against the classical baseline.

Quantum Machine Learning MARL DQN PyTorch Python PQC (Parameterized Quantum Circuits)

VP of Technology

caseHACKS

Sep 2025 - May 2026 (8 months)

Shipped the platform running a 200-hacker event, spanning an organizer control plane (Next.js, Supabase) and participant portal (React, Flask REST API, JWT auth) with constraint-based team matching and realtime QR check-in. Built a sponsor dashboard with live event metrics, filtering, and bulk signed-URL resume exports.

Next.js Supabase React Flask REST APIs JWT QR Check In Event Management

AI/ML Software Engineer

Kinara Systems

Jan 2026 - Apr 2026 (3 months)

Deployed, benchmarked, and fine-tuned Qwen 2.5 3B on edge hardware, improving telemetry and log classification by 10% over baseline while cutting external LLM API cost and keeping customer data on-device. Built a self-hosted OpenAI-compatible inference service for on-device SLMs behind an Apache reverse proxy with API key authentication.

Qwen PyTorch Hugging Face Transformers PEFT Quantization Docker GCP Cloud Run fastAPI REST APIs Qdrant Semchunk Apache API Authentication

Machine Learning Engineer

Dawson Partners

May 2025 - Aug 2025 (3 months)

Built a LangSmith evaluation system with retry logic and multimodal support that invokes LangGraph agents in parallel, making benchmarking fast and repeatable across environments and raising output reliability 33% on time-sensitive financial calculations. Developed a community prompt gallery with FastAPI and deployed the first custom MCP Excel server on GCP Cloud Run with Terraform.

Langsmith LangGraph fastAPI GCP Cloud Run Terraform MCP Excel Docker Python

Junior ML/AI Engineer

IGS

Sep 2024 - Dec 2024 (3 months)

Increased RAG accuracy by 13% and ROUGE scores by 6% with an auto-tester built to score retrieval quality. Achieved 80% classification accuracy across user story domains, platforms, and software stacks via exploratory data analysis and experiments with classical ML and Hugging Face models.

RAG Hugging Face Transformers scikit learn Pandas NumPy Python SQL EDA