Data Science · University of California, Berkeley
Emphasis in Applied Mathematics & Modeling.
Relevant Coursework: Data Structures, Algorithms, Discrete Math, Abstract Linear Algebra, Probability, AI/ML Algorithms, OS
Krish Arora
Software engineer
San Francisco, CA 🌉
Currently building at Zipline and studying Data Science at UC Berkeley. Interested in intelligent systems and the real world they operate in.
About me
Hi! I’m an undergrad at UC Berkeley studying Data Science and Math, with a particular interest in product-facing engineering and backend systems for AI/ML applications. Software and technology are constantly evolving, and I’m excited to be part of shaping what comes next, especially by working on technically challenging problems that bridge powerful backend infrastructure with products people can actually use.
I’ve been fortunate to explore a wide range of engineering problems, from UI/UX design for Qualcomm, workflow infrastructure at AWS, and agent optimization for IBM. Currently, I’m taking a semester off school to intern at the startup Zipline, where I get to develop MC/ARS simulations and validation tools for autonomous drone flights.
Education
Languages
Frameworks & tools
Experience
Software Engineer Intern · Zipline (opens in a new tab)
Currently working on the Autonomy Validation team, where I’m developing flight simulations in C++ and Rust to validate how autonomous drones respond to failures and unexpected route conditions. Alongside this, I’m building a web tool that runs Monte Carlo/Air Risk simulations and validates each run, making it easier to test failsafe behaviors across a wide range of scenarios before they occur in real-world flights.
- C++
- Go
- Rust
Software Development Engineer Intern · Amazon Web Services (opens in a new tab)
Returned to AWS Quick! This summer, I worked more on the infrastructure side, designing and implementing an event-driven notification system that surfaces silent permission and preference changes for over 300K live users without adding latency to user-facing requests. I designed the pipeline to safely handle concurrent processing and retries without sending duplicate notifications, while filtering irrelevant database changes before they reached our Lambda workers. I took the project through its full lifecycle, from writing integration tests and setting up on-call alarms and dashboards to shipping the feature to production.
- TypeScript
- Java
- AWS Lambda
- DynamoDB
- SNS
- ESM
- S3
Software Engineer Intern · IBM (opens in a new tab)
At IBM, I worked on improving how a multi-agent Db2 assistant determines where to route a user’s request. I designed a hierarchical LangGraph system that coordinated eight specialized agents and reduced incorrect tool calls by 60%, then built a hybrid search pipeline using Redis, MiniLM embeddings, and TF-IDF re-ranking that reduced unresolved documentation queries by 40% across 74 indexed sources.
- Python
- LangChain
- LangGraph
- Redis
- Hugging Face
Software Development Engineer Intern · Amazon Web Services (opens in a new tab)
At AWS, I worked on Amazon Quick Flows, which lets users build mini applications connected to enterprise tools to automate their day-to-day tasks. My project focused on expanding its file processing system to support mixed-media uploads. I built an asynchronous pipeline with S3, Lambda, Amazon Transcribe, and DynamoDB that extracted file content so the agent could generate workflows and connect useful tools, such as Jira and Confluence. I also updated the UI to surface transcription progress asynchronously, reducing alert latency by 70% and supporting recovery from failed jobs.
- Java
- TypeScript
- AWS Lambda
- DynamoDB
- Transcribe
- SQS
- S3
Projects
Research · Lead researcher
Tensor Decomposition for DNN Compression
I pursued research on tensor-train decomposition techniques to compress deep neural network layer weights to be used in edge devices, ultimately achieving 87% parameter reduction with little to no performance loss across different network configurations like VGG-16 and ResNet-50. I first-authored the resulting research paper and presented my work to 100+ Ph.D. candidates at UCSB!
- Python
- TensorLy
- TensorFlow
- PyTorch
Google × Qualcomm Developer Hackathon
Settle
I was accepted into the Google × Qualcomm Developer Hackathon, where I built Settle, a contract analysis tool that uses gemma-4-E2B-it-litert-lm and Qualcomm’s Hexagon NPU to detect risky clauses directly on-device. Users can then explore those risks through summaries and follow-up Q&A without sending their documents to the cloud.
- Kotlin
- LiteRT-LM
- OCR
Web application
Aurabase
Aurabase is a social platform I built for friend groups to log events and compete on a dynamic “aura” leaderboard, growing to 250+ users within UC Berkeley. I developed the frontend and backend using React, Express, REST APIs, and Supabase for authentication and data storage.
- React
- Express
- REST APIs
- Supabase
Qualcomm · Contract Data Consultant
Qualcomm Academy
I designed a high-fidelity, interactive Figma prototype for Qualcomm Academy, connecting landing pages, course navigation, and course details. I also prototyped a career quiz that recommends course roadmaps based on users’ interests. The team’s designs were implemented on the Qualcomm Academy website.
- Figma
- UI/UX Design
- Interactive Prototyping
Novo Nordisk · Contract SWE
GLP-1 Review Analysis
For Novo Nordisk, I built an autonomous web scraper to collect 12,000+ user reviews of GLP-1 medications, then developed a docETL-driven NLP pipeline using sentiment analysis and Latent Dirichlet Allocation to identify 8+ recurring adverse-effect topics. Within the collected reviews, the analysis found 35% fewer negative mentions of nausea and fatigue for Novo’s GLP-1 drug.
- Selenium
- Playwright
- docETL
- NLP
- LDA