Projects that show applied AI, product sense, and engineering range.

Architecture

The application uses Next.js, React, TypeScript, and Node.js, with a locally hosted LLM through Ollama. One of the key design decisions was a stateless memory approach: instead of relying on server-side session persistence, the client sends full conversation history with each request so the model can respond with multi-turn context while the system stays simpler to reason about.

Impact

On the interface side, I built a custom design system rather than depending on external component libraries. The result was a responsive, animated chat UI with controlled spacing, visual rhythm, and interaction feedback that made the product feel intentional rather than generic.

• Real-time conversational UX with cleaner message flow and better readability
• Custom CSS variable system and keyframe-based motion without a third-party UI kit
• Component structure designed to keep the codebase maintainable as features expand

Stack

MemoryChat combines front-end product quality with local AI integration, which makes the stack choice central to the project’s identity.

Architecture

I built rule-based checks on top of Python AST parsing to detect undefined variables, bare exceptions, and unsafe default arguments. Using AST-level logic kept the detection process more structured and reliable than surface-level string matching.

• AST-driven parsing for structural analysis of Python code
• Focused checks for real bug-prone patterns that affect maintainability and safety
• Clear separation between detection logic and downstream scoring

Impact

To make the results more actionable, I trained a Logistic Regression classifier to estimate issue severity. That layer made the project more than a linter: it became an example of an engineering tool that uses ML to prioritize attention inside a practical workflow.

The backend was organized with FastAPI and a modular structure that cleanly separated rule-based analysis from model inference, which makes the system easier to maintain, test, and extend.

Stack

AI Bug Finder is built like an engineering utility: strong parsing foundations, ML-assisted prioritization, and a backend designed to stay modular as checks expand.