I'm curious how AI's novelty and intelligence can work under real constraints and actually hold up in the messy, real world.

Figuring out how things fit together is also what drew me to engineering. I'm an experienced chemical engineer - I've delivered oil & gas projects at EPC firms and consulted independently in renewables. And I'm realizing how much of the mindset carries over to building AI through system design, problem decomposition and iterative development.

Here's a project where those ideas come together in practice:

Some of the best experiences I’ve had are trivia nights, yelling out the answer to a random, obscure question and winning!🎉 Online quizzes though are mostly shallow and form-like that don’t quite bring that magic to life.

So how can we create that experience? The recipe would be intriguing questions, the kind that make even a fan go 'wait, I never connected that before...' in a fun, interactive space that is true to canon.

Now the challenges:

• The ability to interpret and judge a free-form answer from the player and make sure it is factually correct, with all the inconsistencies and paraphrasing that comes with it.
• Have a consistent bank of high-quality questions. They would need to be unique across games, genuinely interesting, and grounded in the core knowledge.

How it is delivered: A smart trivia built as an end-to-end AI system. It builds and validates its own dataset from source text offline, then serves it through a fast, interactive Q&A session that uses an LLM only where it's needed. → Explore the repo

✨ The MVP is live and when a group of 12-14-year-olds got their hands on it, they played 8 sessions back-to-back before anyone could pry it away. Early and anecdotal, but a good sign from the toughest critics! 😄

Tech: Python · PyTorch · SBERT · Pydantic · Prefect · Docker · GCP · pandas · NLP / LLMs

Currently applying for roles in applied AI, ML engineering, and data science.

‣Let's connect: Linkedin 💬