Language models
& AI agents
How can agents choose the right tools, retrieve relevant knowledge, and complete tasks reliably? I explore retrieval, skill selection, and the evaluation of multi-step agent behaviour.
Artificial intelligence
Frederick Blais.
Research & engineering
From models to real-world systems.
Profile
AI Researcher
University of Ottawa
Making intelligent systems useful, efficient, and reliable.
I work at the intersection of generative AI and engineering, with a focus on language models, autonomous agents, and AI systems that solve practical problems.
- Focus
- Applied AI & intelligent systems
- Perspective
- Research informed by engineering
- Languages
- English & French
An engineering perspective
Bridging research
and real-world use.
My background in electrical engineering, industrial reliability, and automation shapes how I approach AI research. I am interested in what happens beyond a model's benchmark score: how it behaves inside a working system, what resources it needs, and whether its results can be trusted.
At the University of Ottawa, my research interests centre on generative models and agentic systems. My hands-on work spans retrieval and tool selection, local AI inference, time-series modelling, and reinforcement learning. Across these areas, I favour reproducible experiments and clear, measurable evaluation.
Questions that guide my work
Research interests
Language models
Efficient &
local inference
What makes capable AI practical on accessible hardware? I investigate model quantization, memory use, context limits, and the trade-offs between model quality and inference speed.
Industrial AI
& time series
How can models capture the behaviour of complex industrial processes? I am interested in time-series forecasting, energy modelling, and decision support grounded in engineering knowledge.
Learning through
interaction
What has an agent actually learned? I use interactive environments to study reinforcement learning, reward design, and the gap between apparent progress and robust behaviour.
From questions to experiments
Work & perspective
Evaluating how agents find the right skill
Comparing lexical retrieval, embedding-based search, and reranking for tool and skill selection. The emphasis is on relevance, coverage, latency, and transparent evaluation as skill libraries grow.
Methods
BM25 · Dense embeddings · Reranking · Reciprocal rank fusion
Understanding the limits of local AI
Hands-on evaluation of open-weight language models on memory-constrained hardware. Exploring the relationship between quantization, context length, GPU memory, and practical generation speed.
Methods
Quantized inference · VRAM profiling · Context evaluation
Connecting process data with predictive models
Exploring transformer-based approaches for industrial time series, with an interest in fine-tuning, process behaviour, and the evaluation of predictions against meaningful engineering baselines.
Methods
Time-series transformers · Fine-tuning · Baseline comparison
Useful questions. Testable answers.
I approach AI as an experimental engineering discipline. A promising result needs a clear baseline, an explanation of its limits, and a path to reproduction.
- Define the behaviour
Start with the task, the constraints, and what successful performance would actually look like.
- Measure what matters
Evaluate task success alongside latency, memory use, cost, and failure modes.
- Make the experiment reproducible
Keep code, configurations, evaluation data, and results together so conclusions can be checked.
- Learn from the failures
Investigate unexpected behaviour and distinguish model capability from shortcuts in the environment or evaluation.
Applied experience
Engineering roots.
Research curiosity.
My perspective comes from working with real systems: equipment that must stay reliable, processes that consume energy, and software that needs to behave predictably.
- Electrical engineering
- Electrical projects, industrial maintenance, reliability, and automation.
- Applied computing
- Python, Linux, embedded systems, GPU inference, and practical experimentation.
- Research direction
- Generative AI, language models, agent evaluation, and industrial applications.