Hi, I'm Sharaheel — or Shar for short.
I'm a Computer Science student at the University of Illinois Chicago building AI systems people can trust — with security as the lens I bring to every build. This portfolio documents what I'm building, how I verify it, and where I'm headed.
I learn by building.
I build things to see if they work, not because I'm sure they will. Timely AI happened because I kept forgetting my own schedule and got tired of it. Mate in 10! happened because every chess app I found online was either paid or ugly. This site happened because I wanted something that actually felt like mine instead of a stock theme. None of it was planned. I just kept fixing whatever broke next.
Focus under pressure.
Outside of code I'm probably playing chess, CS, Valorant, or Fortnite, depending on my mood. I'm 1700 Elo and placed at state, which I'm still kind of proud of. The rest is just me being competitive with my friends and losing more than I'd like to admit.
A clear target.
I'm aiming at AI engineering, and I'm doing it the slow way. Right now that means building a tokenizer from scratch so I actually understand what's happening under the hood instead of importing a library and hoping. Every posting I read wanted vector databases, cloud deployment, and orchestration frameworks I don't have yet, so that's next. My security background still shows up in how I think about this stuff, even if it's not the main focus anymore.
Education, skills & experience.
University of Illinois Chicago
B.S. Computer Science, College of Engineering
GPA 3.76 · Dean's List · Expected May 2029
Languages
Frameworks & Tools
AI Engineering & Security
Teach 20+ students Python, JavaScript, Unity, and game dev, walking them through debugging, logic errors, and object-based design until it clicks.
Selected as a Top 40 finalist from nearly 500 applicants. Shipped an open-source PR to WordPress/agent-skills at WordCamp US 2026, published an AI ethics framework, and rebuilt this portfolio as the program capstone.
Built Excel and Sheets dashboards tracking sales, costs, margins, and pricing, then consolidated menu and delivery-platform data to speed up reporting for the owner.
Analyzed sales and inventory data to flag product trends and support purchasing, while running POS transactions and restocking on the floor.
Technical Projects & AI-Built Work
A focused collection of AI, software, WordPress, and interactive app projects that show how I build, verify, and ship practical tools.
Timely AI Calendar Assistant
Timely AI is an agentic scheduling assistant I built with React and Flask. It connects to Google Calendar, classifies every request into one of five intent types, runs the result through a deterministic validation layer, and confirms with the user before anything is written to their calendar.
I proved it works with a 42-case eval harness scoring 98% overall — including driving cancel-intent accuracy from 33% to 98% after reworking intent handling. This project demonstrates agentic architecture, eval-driven development, and user-focused design.
What this project shows
- Five intent types with deterministic validation
- Confirm-before-write calendar safety
- 42-case eval harness — 98% overall
- Cancel intent: 33% → 98% after an intent-handling rework
- AI chat panel for natural-language scheduling
Why this project matters
The demo version of an AI assistant is easy; the hard part is trusting it with someone's live calendar. Timely AI is where I proved I could do that — intent classification, deterministic validation, confirm-before-write, and a scored eval harness that turned "it seems to work" into numbers. It's the clearest evidence of how I want to build AI systems.
Problem
Scheduling can feel scattered when users have to search events or create meetings manually.
Solution
An agentic assistant that classifies intents, validates deterministically, and confirms before writing.
Verification
A 42-case scored eval harness measured real behavior — 98% overall, cancel intent 33% → 98%.
What I Learned
Evals beat vibes — scoring the system is how I found and fixed the cancel-intent failure.
Aim Trainer Reaction Game
This Aim Trainer is a simple reaction-based desktop app I built to help users practice speed, accuracy, and consistency. The user clicks randomly placed targets while the app tracks points, streaks, average reaction time, and time remaining.
I kept the design intentionally clean and focused. The app includes a settings menu where users can switch between Simple and Stats mode, choose 30-second, 60-second, or Infinite play, and pause the game without losing the visual style of the app.
What this project shows
- Interactive target-clicking game logic
- Points, streaks, and average reaction tracking
- Simple and Stats display modes
- 30-second, 60-second, and Infinite timer options
- Clean settings menu and pause overlay
Problem
Reaction games need to feel fast, readable, and uncluttered so users can focus on performance.
Solution
I built a clean aim trainer with targets, scoring, streaks, timers, and a simple settings menu.
Design Choice
I kept the interface minimal so the game feels easy to understand and quick to play.
What I Learned
This helped me practice interactive UI logic, state management, timing, and user-focused design.
Mate in 10! Chess Puzzle Minigame
Mate in 10! was my CS 111 final project, built with Python Turtle. The game starts with a Learn Mode that teaches chess piece movement, then moves into a 10-puzzle practice run where users solve tactics from easy to hard.
My group earned an exemplary grade on this project. It shows my ability to build a larger interactive Python program with visual design, event handling, game logic, feedback, scoring, timing, and a polished user flow.
What this project shows
- Interactive Python Turtle interface
- Learn Mode for chess movement practice
- 10-puzzle run with tactics and difficulty progression
- Timer, misclicks, clean solves, and best-run tracking
- Clear feedback screens and replay options
Problem
Chess tactics are harder to learn without interactive practice, feedback, and progression.
Solution
We created a minigame that teaches movement and then challenges users through 10 puzzles.
My Role
I helped shape the interface, gameplay flow, puzzle structure, debugging, and polish.
What I Learned
This strengthened my Python, event handling, UI layout, testing, and project organization skills.
Portfolio Website Iteration
This before-and-after comparison shows how I used WordPress and AI-assisted iteration to make my portfolio cleaner, more organized, and more professional.
Before, the site had the basic information, but the layout felt plain and harder to scan. After using WordPress customization and AI-supported feedback, I improved the visual hierarchy, navigation, spacing, buttons, theme consistency, and overall presentation.
What changed
- Cleaner navy visual theme
- More professional homepage layout
- Clearer navigation and buttons
- Stronger personal branding
- Better alignment with AI Leaders portfolio expectations
Problem
The original site was functional but looked plain and did not fully communicate my goals.
Process
I used AI feedback to plan improvements, then applied the changes myself in WordPress.
Result
The final version feels cleaner, more professional, and easier for reviewers to navigate.
Skills Shown
WordPress design, page structure, visual hierarchy, content organization, and iteration.
AI Verification Workflow
A workflow showing how I use AI responsibly by checking claims, comparing sources, revising outputs, and using AI as a support tool instead of a shortcut.
Skills shown: AI ethics, verification, research, reflection, responsible AI use, and clear communication.
View AI Workflow →Resume & Technical Direction
My resume and portfolio connect my Computer Science background with AI engineering, agentic systems, security fundamentals, and responsible AI-assisted work.
Skills shown: communication, career alignment, project explanation, and professional presentation.
View Resume →From lessons to proof. My AI Leaders map.
AI Leaders pushed me past "using AI" into working with it responsibly. My workflow became verification-first: I brainstorm and build fast with AI, then slow down to check the important details — testing behavior, reading documentation, and confirming claims before I trust the output.
Across the cohort I produced five artifacts that build on each other — Foundations work that mapped my tools and online identity, AI-leadership pieces on ethics and tool integration, and a career-readiness plan tied to a real Security Analyst path.
The throughline is the same habit I want in AI engineering work: accuracy over speed, and a human staying in the loop on every decision.
The full map walks through every artifact, PDF, and reflection — with the original screenshots and downloadable evidence.
Explore the full AI Leaders Map →AI is a thinking tool, not a judgment replacement.
I use AI to think and document more clearly, never to replace my own judgment. I care about accuracy over speed, which means the work isn't done until I've checked it.
How I Use AI
- Brainstorm and explore approaches before building
- Organize messy notes and next steps
- Clarify concepts I'm still learning
- Improve how I communicate and explain my work
Verification in Practice
In Timely AI, a user types a request in natural language and the app turns it into a Google Calendar event. Before anything is created, it checks for missing details and asks the user to confirm — AI proposes, the human decides.
Why This Matters for My Path
AI engineering depends on not trusting things blindly. The same habit — check the claim, confirm the detail, keep a human in the loop — is how I build systems worth trusting.
The Loop
Brainstorm → Build → Verify. I move fast with AI in the first two steps, then slow down to confirm important claims through documentation, testing, logs, or my own reasoning before I trust the output.
Let's connect.
I'm open to connecting about AI engineering, agentic systems, and internship or early-career opportunities.
What I build, and where I'm headed.
I build AI systems people can trust with their real data. The demo version of that is easy: point a language model at a task and it'll usually look impressive. The hard version is getting it to behave correctly when it's writing to someone's live calendar. That's the work I care about.
Timely AI is where I proved I could do it. It's an agentic scheduling assistant — React on the front end, Flask on the back. Instead of letting the model free-form a reply, I have it sort each request into one of five intent types, and then a deterministic validation step checks that output before anything gets written. It catches scheduling conflicts, and it won't commit a change until the user confirms. To see whether the whole thing held together, I built a 42-case scored eval harness. It came in at 98% overall, and once I reworked how intents were handled, cancel requests jumped from 33% correct to 98%. I have the eval results to back that up.
I also know what I still need to build. Much of my strongest work has leaned on AI assistance, so right now I'm putting real time into the fundamentals I want to own on my own, mainly the math and statistics under machine learning. My fall courses, Applied Linear Algebra and Applied Statistical Methods, go straight at it. And since I've built most things solo, I'm getting fluent with team version control and code review on GitHub now, before it counts on the job.
What I want next is an AI engineering internship for Summer 2027, ideally at a place shipping real AI products rather than a chatbot bolted onto an existing app. After I finish my CS degree at UIC in 2029, I want to do this full-time: building AI systems people can depend on.