class YusraBatool:
def __init__(self):
self.role = "ML Engineer"
self.university = "CS @ Sukkur IBA University '28"
self.location = "Pakistan"
self.focus = [
"Multi-Agent AI Systems",
"Medical Imaging and Explainable AI",
"Cloud-Native ML Pipelines",
"Responsible AI Design"
]
self.current_work = {
"building" : "Agentic AI systems with LangGraph and Google ADK",
"deploying" : "Production ML on Google Cloud Run",
"researching" : "Knowledge graphs for organizational cognition",
"exploring" : "Explainable medical imaging with Grad-CAM"
}
self.philosophy = "Ship fast, ship responsibly, ship things that matter."
def daily_routine(self):
return ["coffee", "research", "build", "deploy", "repeat"]I started with data analytics, moved into machine learning, and found my place building multi-agent AI systems that solve real problems for real people. I care about the full loop — from research to cloud deployment to responsible design. Every project I build goes through the same filter: does it ship, does it work, and does it do right by its users.
Right now I am deep into agentic AI architectures, medical imaging pipelines, and knowledge graph systems. I build with Python, deploy on Google Cloud, and think carefully about guardrails, PII redaction, and human-in-the-loop design in everything I put into production.
I pick tools based on what the problem needs, not what is trending. PyTorch for anything that needs custom model control. Google ADK and LangGraph when orchestration complexity demands a graph-based agent framework. FastAPI when I need a backend that ships today. Cloud Run when I want zero-ops deployment. Neo4j when relationships matter more than rows.
|
Building graph-based memory and multi-agent risk surfacing for teams that need to think together at scale.
|
|
Extending Pneumo-Scan with multi-class pathology detection and investigating attention-based explainability methods beyond Grad-CAM. Working toward models clinicians can actually trust.
|
|
Designing reusable patterns for PII redaction, prompt injection detection, output validation, and regression testing across multi-agent pipelines. Safety is not a feature — it is the architecture.
|
I believe the best AI engineers are not just model builders — they are system designers. An agent that hallucinates costs more than one that says "I don't know." A pipeline without guardrails is a liability, not a feature. I think about failure modes before I think about demos.
Explainable medical imaging system for chest X-ray classification. Uses DenseNet-121 with Grad-CAM visualizations so clinicians can see exactly why the model made its prediction. Built because black-box diagnosis is not good enough for healthcare.
|
Android overlay and multi-agent AI safety assistant designed for elderly Pakistani users. Built in 7 days for the USAII Global AI Hackathon. Placed Top 75 out of 424 teams globally.
|
|
Multi-agent job application workflow with persistent tracking and memory. Featured in Google Cloud Builder Stories for demonstrating production-grade agentic AI.
|
Organizational cognition engine that combines multi-agent orchestration with knowledge graphs. Agents surface risks, map institutional knowledge, and maintain graph-based memory across sessions.
|
|
Multi-agent medical information assistant with separate research and advisory flows. Designed with safety guardrails so it assists without overstepping clinical boundaries.
|
AI-powered journaling companion. Built and deployed solo in under 24 hours at MLH AI Hackfest 2026. Proof that good tools ship fast when the architecture is right.
|
|
GUI maze game with real-time graph traversal visualization. Implements BFS, DFS, and Dijkstra with animated pathfinding. Recognized at AI and CS Expo 2024.
|
Role-based academic management platform with full CRUD operations, authentication, and relational database integration.
|
|
Command-line vocabulary game built in C++. Clean logic, no frameworks, just algorithms and data structures doing what they do best.
|
|
I do not build projects to fill a portfolio. I build them because I see a gap — a workflow that could be automated, a diagnosis that should be explainable, a user group that technology forgot about. The project list is a side effect of caring about the right problems.
Jun 2025 - Present |
May 30 - Jun 22 |
Jan 3 - Feb 18 |
88% uptime |
avg daily |
I do not chase credentials — but when programs align with what I am already building, I show up and deliver. Here is what that looks like:
|
Selected by Google Cloud to be featured in their Builder Stories program for demonstrating production-grade multi-agent AI deployment. SmartApply was highlighted as an example of how agentic AI systems can be built and shipped using Google Cloud infrastructure — from Firestore for persistent memory to Cloud Run for zero-ops deployment.
|
|
First cohort of Google Cloud's Generative AI Academy for the APAC region. Track 1: Build and Deploy AI Agents. Covered agent design patterns, tool integration, memory architectures, and production deployment on Google Cloud. Applied learnings directly to SmartApply and MediGuide AI.
|
|
Built ElderBridge GuardianOS from scratch in 7 days — a multi-agent AI safety assistant designed for elderly Pakistani users. The system runs as an Android overlay using AccessibilityService, intercepts potential scam calls and messages, and routes them through a FastAPI backend connected to LangGraph agents powered by Azure OpenAI. Placed in the top 75 out of 424 teams globally in the qualifier round.
|
|
Selected as a participant from a highly competitive global applicant pool. The Aspire Leaders Program develops emerging leaders through Harvard-designed curriculum focused on leadership development, community impact, and personal growth. Applied leadership frameworks directly to organizing SIBAthon'26 and leading project teams.
|
|
Built and deployed Matcha Journal completely solo in under 24 hours — an AI-powered journaling companion that uses Groq-hosted LLaMA 3.3 for real-time conversational journaling with mood tracking and reflection prompts. Went from zero to live deployment on Streamlit Cloud within the hackathon window. Proof that the right architecture choices make speed possible without sacrificing quality.
|
|
Completed McKinsey's Forward program — a professional development curriculum covering structured problem-solving, adaptive communication, and leadership in complex environments. Applied these frameworks to how I scope projects, communicate technical decisions, and break down ambiguous problems into actionable engineering plans.
|
|
Awarded under the Prime Minister's Youth Laptop Scheme — a merit-based national program for high-performing students in Pakistani universities. The laptop I received on this program is the same machine I used to build my first AI projects, deploy my first cloud applications, and write every line of code that got me to where I am today. |
|
Received Certificate of Recognition at the AI and Computer Science Expo 2024 for Maze Adventure — a Java Swing application implementing real-time graph traversal visualization with BFS, DFS, and Dijkstra pathfinding algorithms. One of my earliest projects that proved to me that clean algorithm implementation and good UI design are not mutually exclusive.
|
|
Served on the organizing team for SIBAthon'26 — the university's flagship hackathon event. Coordinated logistics, mentored participating teams on AI and cloud deployment strategies, and helped design challenge tracks. Building things is one skill. Helping others build things is another. |
{
"certifications": [
{
"name": "Generative AI Leader Professional Certificate",
"issuer": "Google Cloud",
"status": "verified"
},
{
"name": "Google Data Analytics Professional Certificate",
"issuer": "Google",
"status": "verified"
},
{
"name": "Data Analytics Certificate",
"issuer": "IBM SkillsBuild",
"status": "verified"
},
{
"name": "Excel Essentials for Data Analytics",
"issuer": "IBM",
"status": "verified"
},
{
"name": "McKinsey Forward Program",
"issuer": "McKinsey and Company",
"status": "verified"
},
{
"name": "SQL Intermediate",
"issuer": "HackerRank",
"status": "verified"
},
{
"name": "Java Basic",
"issuer": "HackerRank",
"status": "verified"
}
]
}
Additional completed work: Deloitte Data Analytics Job Simulation, AWS Aurora MySQL Basics, Microsoft AI Skills Fest, Google Arcade Facilitator Program.
[2026-06-22] Currently exploring:
- Advanced multi-agent memory architectures
- Graph neural networks for knowledge representation
- Production observability for ML pipelines
- Android accessibility services for inclusive AI
- Responsible AI evaluation frameworks
[NEXT] On the radar:
- MLOps pipeline automation with Vertex AI
- Federated learning for privacy-preserving medical AI
- Building evaluation harnesses for agentic systems
- Contributing to open-source AI safety tooling
I treat learning the same way I treat engineering — systematically, with clear intent, and always tied to something I am building. Courses are inputs. Shipped projects are outputs. The ratio matters.
yusra@dev ~
-----------
OS : Builder, not just a learner
Uptime : Since 2024, shipping consistently
Shell : Python, Java, Kotlin, C++, SQL, R
Terminal : VS Code + Google Cloud Shell + Android Studio
Packages : 9 projects shipped, 3 live in production
Resolution: Build AI that is useful, explainable, and safe