Skip to content

Full-stack engineer

Mohammad Imrose

Full-stack + AI: React, Next.js, Node, TypeScript, LLM workflows, AWS.

Focus · AI · Full-stack Base · Sacramento

Tooling, shipping real systems.

Scroll

Work

Selected work

Skim the one-liner, then open Work / name for challenge, approach, outcome, impact, and extra detail.

01

CalAssist Mortgage Fund

2026 · OsaaS

SLA reporting for state agencies → faster SQL, aging reports, Lambda + RDS patterns.

  • MariaDB
  • Lambda
  • Node
  • SQL
  • Sequelize
Full story

Challenge

State agency case managers needed timely visibility into disaster-relief mortgage assistance: which applications were aging past SLA, how workloads compared across programs, and whether reporting stayed accurate when federal rules and business calendars (including holidays) changed. The stack mixed Angular and Spring Boot with MariaDB and asynchronous work in AWS Lambda, so performance and observability had to hold up under concurrent agency users.

Approach

I focused on the data layer first: query plans, indexing, and refactoring heavy multi-join SQL so dashboards and aging reports could run without table scans. I built and tuned REST APIs in Spring Boot with validation and consistent error shapes, and moved appropriate work to Lambda with buffered logging and Secrets Manager integration so serverless handlers could talk to RDS through proxy-friendly connection patterns. Where reporting needed cross-program aggregates, I used multi-CTE queries and careful normalization so metrics stayed explainable to stakeholders.

Outcome

  • Average query time dropped ~40% on the hottest paths after schema and query work.
  • Aging and SLA reports gave managers a clear queue of at-risk cases.
  • Lambda-based async work reduced load on synchronous APIs with audit-friendly logging.

Impact

~40% faster queries on targeted routes; fewer production errors from API validation; clearer operational story for state partners reviewing relief metrics.

Details

  • Collaborated with agency stakeholders to encode business-day SLA and holiday rules in reporting logic.
  • Sequelize singleton + secrets patterns aligned with RDS Proxy expectations in Lambda.
02

Schedia Study Buddy (NoteVision)

2025-present

PDFs → study UI → Claude + agentic-doc, SSE (FastAPI↔Next), KaTeX.

  • Next.js
  • FastAPI
  • Claude
  • SSE
  • KaTeX
Full story

Challenge

Students upload long PDFs (lectures, scans, mixed layouts) and expect structured notes, math that renders correctly, and a UI that feels like a desktop workspace rather than a single long chat. Pipeline runs can take minutes; the client must show honest progress without blocking the rest of the app.

Approach

I wired an agentic pipeline with Anthropic Claude and Landing AI agentic-doc to parse PDFs and emit structured content, then layered Vercel AI SDK and KaTeX on the Next.js App Router side for equations and rich text. The Python FastAPI backend streams progress over Server-Sent Events into the browser so users see stage-by-stage updates. For navigation, I leaned into a Windows-inspired metaphor (windows, taskbar, Start menu) so dense study material stays scannable.

Outcome

  • End-to-end flow from upload to rendered study UI with streaming feedback.
  • KaTeX for math-heavy content; structured notes from agentic-doc parsing.
  • Long-running jobs show honest progress, not black-box waits.

Impact

Faster trust in the pipeline (users see what step is running); better support for STEM coursework; foundation for more agent steps without rewriting the transport layer.

Details

  • SSE chosen over polling to reduce load and simplify “live” UX for long-running inference.
  • Agentic-doc reduced manual cleanup of PDF-derived HTML.
03

uni-scraper-pipeline

2026

Auto-pull degree reqs from messy catalogs → GPT-4 loop, Playwright, Postgres.

  • Python
  • Playwright
  • Postgres
  • GPT-4
Full story

Challenge

Universities publish degree requirements in different CMS layouts, often behind JavaScript-rendered pages. Manually collecting every catalog URL does not scale; the system had to discover paths, respect robots and rate limits, and stay within API budgets.

Approach

I implemented a plan-execute-observe loop with GPT-4 and SerpAPI to reason about where catalog content lives, combined with Playwright for pages that need a real browser. Data lands in PostgreSQL via asyncpg with Pydantic v2 models so courses, prerequisites, and requirement groups stay normalized and type-checked. Token usage and per-domain throttling are tracked explicitly.

Outcome

  • Autonomous extraction across heterogeneous university sites from one pipeline.
  • Normalized PostgreSQL storage ready for planners, analytics, or exports.
  • GPT-4 + SerpAPI discovery plus Playwright when the DOM requires a browser.

Impact

Less manual curation of URLs; predictable cost controls; schema ready for Schedia-style planning products.

Details

  • Robots.txt compliance and polite crawl intervals by domain.
  • Playwright handles SPAs that static fetch cannot.
04

GuideCode

2025

Chat → specs + Cursor-sized tasks + UX agent that remembers decisions.

  • GPT-4
  • Product
  • Prompting
Full story

Challenge

Early ideation is messy: stakeholders mix goals, constraints, and UI wishes in one thread. Engineers need something closer to a PRD and a backlog of Cursor-sized tasks, without adopting a heavyweight project suite.

Approach

I used GPT-4 to run a conversational “interview” that converges on JSON specs (architecture, stack, features), then a task-breakdown step that emits prompts sized for AI-assisted IDEs. A separate UX design agent proposes flows and components; drag-and-drop history plus duplicate detection keeps suggestions from repeating.

Outcome

  • Single flow from vague idea → JSON spec → ordered Cursor-sized tasks.
  • UX design agent stays in-session with history and duplicate detection.
  • Specs and tasks stay linked so implementation prompts match the agreed design.

Impact

Shorter path from brainstorm to implementation prompts; fewer duplicate UX proposals as the conversation grows.

Details

  • Tasks are framed so they paste cleanly into Cursor-style workflows.
05

Schedia.ai

2025-present

Planning + study agents → Three.js / R3F shell, motion, glass UI.

  • Three.js
  • R3F
  • LLM
  • Framer Motion
Full story

Challenge

Academic life mixes planning (courses, deadlines) with on-demand help (explain this, quiz me). A flat chat list does not communicate semester context or urgency; the product needed an immersive shell that still works on phones.

Approach

I built agents for semester planning and a real-time Study Buddy that share context about courses and deadlines. The front end uses React Three Fiber and Three.js for depth and spatial interest, with glassmorphism, Framer Motion, and responsive layout so the same ideas read well on desktop and mobile.

Outcome

  • One surface for semester planning, explanations, quizzes, and practice.
  • Three.js / R3F shell with glass UI that reads as a product, not a demo chat.
  • Responsive layout so the same flows work on phone and desktop.

Impact

Clearer mental model for students (plan vs. ask vs. practice); more engaging entry point for the product.

Details

  • Motion used for hierarchy and feedback, not decoration-only.
  • Shared agent context reduces repeated explanations.

Path

Experience

  • Full-Stack Engineer

    Now

    OsaaS LLC · Sacramento, CA

    Jan 2026 - Present

    • Enterprise stack: Angular + Spring Boot with clearer module boundaries for independent scaling.
    • MariaDB: ~40% faster queries via indexing, query plans, and refactors on heavy joins.
    • AWS Lambda for async work; Spring REST APIs with validation and structured errors; SQL tuning under concurrency.
  • Software Engineer I

    Algorizin Inc. · New York, NY

    Aug 2025 - Jan 2026

    • MigrateEasy: React + Node/Express immigration automation; ~30% faster user tasks and fewer manual entry errors.
    • REST APIs for profiles, documents, and workflows; Winston + Sentry for operations.
    • Jest + Supertest; API docs that shortened onboarding for new engineers.
  • Software Engineering Intern

    California Department of Technology · Rancho Cordova, CA · Hybrid

    Dec 2023 - Dec 2024

    • Managed complete project process from planning to testing, ensuring timely and cost-effective delivery.
    • Engineered scalable, maintainable user interfaces with React.js and Next.js; improved application performance by ~30%.
    • Rigorous QA testing and debugging; resolved 50+ critical issues for cross-browser compatibility.
    • Designed and implemented RESTful APIs to streamline data exchange between applications and backend services.
    • Collaborated with cross-functional teams to launch a new ServiceNow platform and improve internal workflow efficiency.

Stack

Skills & tools

Stack I use most. Not an exhaustive list. Use the shortcuts to jump to a group.

01 / 05

Languages

What I write in production code.

  • Java

  • Python

  • SQL

  • JavaScript

  • TypeScript

  • HTML

  • CSS

02 / 05

Product & UI

Frameworks and surfaces users touch.

  • React

  • Next.js

  • Angular

  • Node.js

  • Express

  • Vite

  • MUI

  • Framer Motion

03 / 05

AI & agents

Models, SDKs, and streaming patterns.

  • Claude

  • GPT-4

  • Vercel AI SDK

  • LangChain

  • SSE

  • agentic pipelines

04 / 05

Data & cloud

Persistence, infra, and delivery.

  • PostgreSQL

  • MariaDB

  • MongoDB

  • Firebase

  • Supabase

  • AWS

  • Docker

  • GitHub Actions

05 / 05

Quality

Tests, types, and observability.

  • Jest

  • React Testing Library

  • Playwright

  • Pydantic v2

  • Sentry

  • Winston

About

About me

Background

Full-stack engineer focused on AI-assisted products and developer tooling. I ship end-to-end with React, Next.js, Node, and TypeScript, and I integrate LLM pipelines, agentic workflows, and AWS (Lambda, data stores, CI) where reliability and observability matter. I care about clear APIs, honest loading states, and systems that stay understandable when requirements shift.

Hello

Let’s talk

Open to roles: AI or strong product engineering teams.