Behavioral Interview Practice

AIMING Lab — Situation

In Spring 2025, I joined UNC’s AIMING Lab to work on a project analyzing how GPT-2 makes predictions. I was asked to build a token attribution tool — using PyTorch, which I had never used before.

AIMING Lab — Task

I had just two weeks to learn PyTorch from scratch and develop a fully functional tool that could analyze and visualize which words influenced GPT-2’s output the most.

AIMING Lab — Action

• Created a focused learning plan using tutorials, docs, and GitHub code.

• Built and tested a working prototype using gradients for attribution.

• Optimized the system and deployed it to a Tesla T4 GPU with batch processing.

AIMING Lab — Result

The tool ran nearly 5x faster — from 2.17s to 0.44s per input — and was successfully used in real research experiments.

AIMING Lab — Impact

The tool was adopted by other researchers in the lab, saving time and helping advance multiple ongoing projects in model explainability.

AIMING Lab — Philosophy

I learned that even complex tools can be mastered quickly through hands-on learning and clear goals — building something real helps everything click faster.

Ember Learning — Situation

While interning at Ember Learning, I disagreed with a teammate on whether to use global or component-specific loading spinners for an AI tutoring platform feature.

Ember Learning — Task

We needed to resolve the disagreement quickly to meet our sprint deadline without compromising performance or user experience.

Ember Learning — Action

• Suggested building two prototypes to compare both approaches.

• Gathered team feedback through simulated user testing.

• Facilitated a design review with our tech lead to finalize a direction.

Ember Learning — Result

We combined both approaches: component spinners for complex views and a global fallback — improving responsiveness while keeping the codebase clean.

Ember Learning — Impact

This decision improved the app and set a precedent for how our team handled future design debates — with testing, collaboration, and compromise.

Ember Learning — Philosophy

I learned that resolving conflict isn’t about being right — it’s about finding the best solution for the team through empathy, evidence, and respect.

Learning Ally — Situation

At Learning Ally, I joined a cross-functional team tasked with building a new platform to help over 8,000 educators across 400+ districts manage literacy assessments.

Learning Ally — Task

My job was to build backend systems and ensure smooth integration with external tools while also collaborating with designers and educators.

Learning Ally — Action

• Developed 7 RESTful APIs for syncing assessment data.

• Worked with 5 designers to align UI with classroom workflows.

• Monitored usage and improved performance in weekly sprints post-launch.

Learning Ally — Result

The platform launched successfully, usage doubled during the pilot, and I received the Mary Alyce and John Merow Over-Achiever Award.

Learning Ally — Impact

The platform helped teachers identify struggling students more efficiently and increased district-wide engagement with assessment tools.

Learning Ally — Philosophy

I learned that effective software development means constant iteration, user empathy, and strong teamwork — especially when building tools that impact education.

EngageSTEM — Situation

I founded EngageSTEM, the first student-led STEM nonprofit in my hometown, to give students hands-on STEM experiences and mentorship.

EngageSTEM — Task

I needed to build the website, develop a matching platform, and lead outreach efforts to grow the program from the ground up.

EngageSTEM — Action

• Built the website using React and Django.

• Created a mentor-matching system to connect students with tutors.

• Led community outreach that boosted attendance by 50%.

EngageSTEM — Result

The platform successfully matched dozens of students and mentors, and community participation and retention increased significantly.

EngageSTEM — Impact

EngageSTEM gave local students access to STEM education they wouldn’t have had otherwise, and established a scalable, repeatable nonprofit model.

EngageSTEM — Philosophy

I learned that leadership means building systems that empower others — and that meaningful impact often starts at the local level.

Tell Me About Yourself — Intro/Background

I’m Digraj Pelia, a sophomore at UNC-Chapel Hill majoring in Computer Science with a minor in Statistics & Analytics. I’ve always been passionate about building things that solve real-world problems, especially at the intersection of education, AI, and software.

Tell Me About Yourself — What I’ve taken on

Over the past two years, I’ve actively sought out internships and research experiences where I could contribute to meaningful projects, improve my technical skills, and learn how software makes an impact at scale.

Tell Me about Yourself — Action (What I’ve done)

• At Ember Learning, I helped develop an AI tutoring assistant that now supports over 5,000 users.

• At UNC’s AIMING Lab, I built a PyTorch tool that helps researchers understand how GPT-2 makes predictions.

• I also founded EngageSTEM, a nonprofit where I built a tutoring platform and led outreach that grew student participation by 50%.

Tell Me About Yourself — Result

These experiences have taught me how to work with diverse teams, ship real products, and create technology that’s both innovative and user-centered.

Tell me About Yourself — Philosophy

Right now, I’m looking for internship opportunities where I can grow as an engineer, work on technically challenging problems, and learn from mission-driven teams solving real problems.