Hafsah Nasir

Human-Centered AI Architect | CS @ LUMS | Building Scalable, Bilingual AI Tools

Overview

Computer Science student at LUMS specializing in Human-Centered AI and scalable system design. From leading Pakistan’s largest science olympiad (PsiFi) to building bilingual AI learning companions for low-SES classrooms, I bridge the gap between complex technical logic and real-world user needs. Experienced in ML automation, full-stack development, and leading high-impact teams.

Lahore, Pakistan

Experience

Jul 2025 – 2026

VOLUNTEER

Convenor

PsiFi XVII – SPADES, LUMS

Lahore, Pakistan

Led Pakistan’s largest national science olympiad with 1,500+ participants, 120+ institutions, and an 80+ member organizing team. Implemented crisis-management systems and workflow automation, delivering record engagement and 35,000+ online reach.

•Led Pakistan’s largest national science olympiad with 1,500+ participants, 120+ institutions, and an 80+ member organizing team.

•Implemented crisis-management systems and workflow automation, delivering record engagement and 35,000+ online reach.

Key Achievements

→Delivered record engagement and 35,000+ online reach.

Workflow AutomationCrisis-management systems

2025 – 2025

WORK

Machine Learning Intern

Orbin

Lahore, Pakistan

Built an AI-assisted automation pipeline to stitch multiple drone-captured panorama images into a single high-resolution composite, applying image alignment, feature matching, and stitching logic. Improved robustness for noisy drone imagery by handling overlap inconsistencies and perspective variation, applying computer vision and ML concepts to a real-world geospatial imaging problem.

•Built an AI-assisted automation pipeline to stitch multiple drone-captured panorama images into a single high-resolution composite, applying image alignment, feature matching, and stitching logic.

•Improved robustness for noisy drone imagery by handling overlap inconsistencies and perspective variation, applying computer vision and ML concepts to a real-world geospatial imaging problem.

Key Achievements

→Built an automated pipeline to align and merge batches of ~20 overlapping drone images into seamless high-resolution composites with ~90% stitching accuracy.

→Strengthened pipeline robustness against exposure differences, perspective distortion, and overlap inconsistencies using feature detection, keypoint matching, and homography-based warping.

→~90% stitching accuracy on batches of 20+ images

Computer VisionFeature DetectionKeypoint MatchingHomography-based WarpingImage StitchingPython

Oct 2023 – Feb 2024

VOLUNTEER

Lead Coordinator

Startup Weekend Lahore

Lahore, Pakistan

Led a 30-member team to execute a 54-hour Startup Weekend in collaboration with Google for Startups and Techstars. Directed branding and program flow, achieving 2× social media growth and a seamless, high-impact participant experience.

•Led a 30-member team to execute a 54-hour Startup Weekend in collaboration with Google for Startups and Techstars.

•Directed branding and program flow, achieving 2× social media growth and a seamless, high-impact participant experience.

Key Achievements

→Achieved 2× social media growth.

Branding

Project Portfolio

ACADEMIC

Present

Suicide Rate Analysis & Prediction (Data Science)

Data Analyst

Analyzed 30+ years of data across 180+ countries (WHO, World Bank, UNDP HDI) to test how economic crises affect suicide rates, building regression models and running hypothesis tests in Python (pandas, NumPy, statsmodels, scipy). Found high inflation a significant predictor of rising rates two years later (p = 0.003).

PythonpandasNumPystatsmodelsscipy

PERSONAL

Present

Cosmos Discovery – Campus Event Platform (Android)

Lead Backend & Recommendation Engine Developer

Owned the event services and Discover feed, implementing event creation/management and a real-time RSVP flow over Firestore snapshot listeners with optimistic updates and rollback on failure. Designed and built the content-based recommendation engine that ranks events by category and tag overlap with each student's RSVP history, with exclusion rules and a popularity-based fallback. Backed by JUnit and Espresso tests and a GitHub Actions CI pipeline.

AndroidJavaFirebase AuthFirestoreCloud FunctionsJUnitEspressoGitHub Actions

Code

PERSONAL

Present

Seekho Sikhao – AI Learning Companion (HCI, CS466)

Lead Designer & Developer

Led the human-centered design lifecycle for a bilingual (Urdu/English) AI tutoring tool for low-SES Pakistani classrooms, running focus groups, teacher interviews, and task-based field sessions, then translating findings into system requirements. Designed the UI/UX, including voice-enabled Urdu responses and a split-screen workbook that prompts students to attempt problems before revealing scaffolded hints, iterating through paper, Figma, and React prototypes. Built the Wizard-of-Oz testing prototype (TypeScript/React, deployed on Vercel), driving development with AI coding tools (Claude Code, Codex) via structured Markdown specs and context/memory management.

TypeScriptReactVercelFigmaClaude CodeCodexMarkdownAI Coding Tools

Live Code

PERSONAL

Present

ThriftHub: Database-Backed Marketplace System

Database Designer & Backend Developer

Designed and implemented a complete relational database system for a campus thrift marketplace, including ER diagrams and schema normalization. Developed the backend logic for user management, product listings, transactions, and search functionality.

SQLRelational Databases

Education

2027

Bachelor of Science

LUMSSpecialization in Computer Science

GPA: 3.18

Relevant Coursework

Human-Computer InteractionMachine LearningPrinciples & Techniques of Data ScienceAlgorithmsData StructuresDatabasesOperating SystemsNetwork-Centric ProgrammingNetwork Security