Product Engineer

Problem

Students rarely struggle because they lack content; they struggle because notes, slides, readings, and deadlines are scattered, and they cannot quickly tell what they understand.

Product

A learner uploads course files or web content, organizes it by course and week, then gets grounded chat, summaries, Notes, quizzes, cuecards, concept maps, and adaptive recommendations.

What I built

• End-to-end upload and ingestion flows for multiple content types, chat with content, citations, multi-model, tools and artifacts generation and consumption.
• Retryable background jobs that index uploaded material, track progress, and generate study assets without blocking the user.
• AI generation paths for summaries, quizzes, cuecards, Golden Notes, open questions, concept maps, and grounded chat responses.
• Onboarding, usage quotas and limits, billing, analytics, caching, rate limits, and protected multi-tenant data access.

Problem

A normal chat answer can sound confident even when it lacks fresh context, shows no work, and gives the user no source trail to inspect.

Product

The app signs users in, rewrites the question into search plans, gathers web context, streams each research step, decides when it has enough information, and saves the sourced conversation.

What I built

• An agent loop that plans searches, runs web retrieval, evaluates whether more context is needed, and then writes the final answer.
• Streaming chat UX with action steps, resumable responses, and progress updates so users can follow the research instead of waiting on a black box.
• Authentication, per-user usage limits, global API protection, chat persistence, structured logging, and monitoring.
• Evaluation and observability hooks so answer quality, tool behavior, and regressions can be inspected instead of guessed.

Problem

Full-stack apps often drift: forms validate one shape, APIs expect another, protected flows grow inconsistent, and UI updates lag behind user actions.

Product

A Basic clone of X (Twitter) app demonstrates profiles, experiences, attendance, comments, favorites, tags, search, notifications, image uploads, and protected user flows.

What I built

• Shared Zod schemas for forms and server inputs so the client and API agree on the same validation rules.
• Protected tRPC procedures, JWT refresh flows, output validation, and reusable server middleware for authenticated workflows.
• Optimistic UI updates, cache rollback, infinite feeds, filtering, image uploads, and targeted invalidation.
• A client, server, and shared package structure that lets reviewers trace a feature from route to database and back to the UI.

Problem

A fleet system has several moving parts that should run independently: a web app, API gateway, tracker, simulator, queue, and database.

Product

Kubernetes manifests deploy the full local environment, expose the web UI and queue console, and keep the internal API, tracker, and database private inside the cluster.

What I built

• Kubernetes workloads for the Angular web app, API gateway, position simulator, position tracker, ActiveMQ, and MongoDB.
• Service boundaries that expose only the user-facing web UI and queue console while keeping internal traffic on ClusterIP services.
• Persistent MongoDB storage so vehicle state can survive pod restarts in a local cluster.