Portrait of Sara Jalili Shani, Machine Learning Researcher

Hello, I’m

Sara Jalili Shani

Machine Learning Researcher

Canada · Open to remote opportunities

My background is in game theory, market design, and optimization. I now work on multimodal AI, focusing on inference-time tool selection for vision-language models. Across both areas, I’m interested in how intelligent systems make complex decisions and how their behaviour can be evaluated systematically.

Email me GitHub LinkedIn CV
✦

About me

I completed my B.Sc. at Amirkabir University of Technology and my M.Sc. in Computer Science at the University of Alberta under the supervision of James R. Wright . My master’s research focused on using optimization and simulation to study spectrum auctions. I now work as a Machine Learning Researcher at Algonet, where I investigate recent methods in multimodal AI and computer vision, evaluate models on established benchmarks, and develop algorithms to improve their performance on visual reasoning tasks.

Portfolio

Featured work

5 projects

Ongoing projects

TOOLS

Tool Selection for Multimodal Models

Developing algorithms that select and route external visual tools to improve performance on multimodal reasoning and computer-vision tasks.

Multimodal AI Tool Selection Computer Vision Ongoing
FAIR

Improving Fairness in Conformal Prediction

Ongoing research on improving fairness and reliability in uncertainty-aware machine-learning systems. Further details will be shared as the project develops.

Conformal Prediction Uncertainty Fairness Ongoing

Previous work

VLM

Evaluating Chain-of-Thought Reasoning in Qwen 3 and Gemma 4

Evaluated Qwen 3 Instruct and Thinker models and Gemma 4 with chain-of-thought prompting on MMBench and BLINK through a reproducible benchmarking pipeline.

Qwen 3 Gemma 4 MMBench BLINK
View project ↗
AUC

Simulating Canada’s 3800 MHz Spectrum Auction

Built an optimization-based simulation to study how alternative auction policies affect allocation, deployment, and population coverage.

Optimization Simulation Market Design Spectrum Auctions
View project ↗
GERIS

GERIS: Filtering Label Noise in License Plate Data

Developed a game-theoretic framework for filtering instance-dependent label noise in synthetically augmented license-plate data.

Game Theory Computer Vision Label Noise
View project ↗
⌁

Publications

```