I am a software engineer with a keen interest in developing scalable systems, especially in the domain of machine learning and web technologies. I like making things that facilitate gathering and create a sense of community. I am currently working at Fullscript, where I am building the search stack. Previously, I worked at Zafin, where I led the development of a query engine to enable embedded analytics for banking systems. Prior to Zafin, I worked at Signiant, where I built enterprise applications for file transferring. Beyond coding, I am active in the local tech scene. You can find me in tech events and meetups in town. I am based in Ottawa, Canada.
Tool calling enables LLMs to operate in an expanded environment such as a browser or a coding environment. ReAct and CodeAct offer two distinct paradigms of integrating tool calling for agents. In practice, both of these patterns can be implemented in a variety of ways. Developers will have to weigh the pros and cons of different patterns and implementations as they try to build agents that are both powerful and reliable.
If anything, I think the success of Bolt and Lovable proved that LLM applications can do wonders with prompting and simple architecture. I suspect that this type of application will not disappear any time soon.
I present a few findings and observations in the space of designing and training small LLMs.
Embedded analytics brings analytics as close to business operations as possible, reducing the amount of unnecessary context-switching between application interfaces.
At Zafin, I led the development of a core middleware across two teams of cross-functional teams. The middleware leveraged a query engine to make analytics dashboards for enterprise banking systems. I was responsible for the design and implementation of the middleware with the existing product, ensuring its performance and scalability through k8s design patterns and extensive testing. The success of the project led to the adoption of the same architecture in another team.
My work at Signiant involved fullstack development of its cloud native application. I implementation admin features for managing and monitoring the file transfers. I use Test Driven Development (TDD) to ensure the quality of the code, shipping features with 95% code coverage.
A significant proportion of my time during my master's degree at uOttawa was spent on building robotics. I worked as a research assistant to develop an ethics of AI course, where I hacked robots to simulate ethical dilemmas in human-robot interaction. I was also a member of the uOttawa Mars Rover team, where I developed robotics navigation and camera systems.