Kamran Ahmed
Kamran Ahmed
Software Engineer
Resume

About me

Recently graduated Master's student in Computer Science from Stanford University with a passion for computer networking and systems programming. Experienced in building network protocols, distributed systems, program analysis tools, and cloud-based applications.

kmahmed[at]cs.stanford.edu

Education

MS, Computer Science

Stanford University

September 2022 - June 2024

BA, Molecular and Cell Biology, Neurobiology

University of California, Berkeley

August 2015 - May 2019

Experience

Software Engineer Intern

Capital One

June 2023 - August 2023

Optimized Capital One’s step-up authentication decision engine and fraud risk monitoring system, substantially reducing customer authentications during call sessions. This enhancement in customer satisfaction yielded annual savings of over $3M for real-time fraud decision-making platforms. Built a high-throughput, streaming-based serverless application that efficiently handled Kafka events, sent them to an SQS queue, and processed them asynchronously with Lambda functions. Demonstrated expertise in a wide range of technologies, including AWS (Lambda, Fargate, SQS, IAM, CloudWatch, and DynamoDB), Apache Kafka, Python, pytest, Behave, Splunk, and Jenkins.

Software Engineer Intern

HiHome (Startup)

May 2022 - August 2022

Spearheaded technical development of a real estate workflow management application, architecting backend solutions, including an API service, an event and task manager, and third-party integrations with Google APIs and Follow Up Boss. Enhanced the home-matching platform by optimizing multiple services, resulting in improved frontend usability and enhanced backend API performance, reducing geolocation query times by 85%. Collaborated with a remote development team on HiHome’s FastAPI service and Elasticsearch scoring engine. Applied expertise in FastAPI, SQLAlchemy, PostgreSQL, Redis, and Docker.

Projects

Pintos Operating System

Developed thread management, virtual memory, and file system functionality in C. Implemented system calls to enable interaction between user programs and the kernel.

User Space TCP

Recreated a full-fledged TCP networking stack in user space with C++. Created a traceroute-like network analysis tool to examine how packets are routed through the Internet.

Peer-to-Peer VPN

Created a VPN similar to Wireguard with Rust and C. Encrypted IPv4 traffic with AES-GCM and implemented a sliding window protocol to mitigate replay attacks.

DNS Server

Built a DNS recursive resolver in Rust. Created an LRU cache to optimize query resolution and a custom parser and serializer to minimize data copying overhead.

Sidekick

Built a sidekick protocol in C++ that assists secure end-to-end transport protocols over asymmetric network paths as part of a replication study of the Sidekick NSDI paper. Implemented a selective acknowledgment mechanism to efficiently acknowledge packets without access to cleartext sequence numbers using polynomial power sums and modular arithmetic. Replicated the original study’s results using Mininet and real-world testing, achieving a 51% reduction in de-jitter latency of a low-latency audio stream over a lossy network in emulation (vs. 52% in the study) and a 79% reduction in real-world scenarios (vs. 91% in the study), demonstrating the proxy’s effectiveness across both environments

Catamaran

Engineered a fault-tolerant and distributed DNS nameserver in Go that replicates DNS resource records using a custom Raft implementation. Integrated Dynamic DNS (DDNS) updates to allow services with continuously changing IP addresses to update resource records without manual intervention. Conducted extensive evaluation of query and update latency, fault tolerance, and replication costs compared to BIND 9. Found that while Catamaran had a low cost of replication for reads, it incurred a significantly higher cost for writes due to the sequential processing of DNS updates forwarded to the Raft leader, resulting in greater latency compared to BIND 9.

tainted

Developed a program analysis tool for Python to dynamically identify and track the flow of sensitive data in consumer-facing applications. Implemented dynamic taint tracking through program instrumentation and a custom runtime library, minimizing interference with normal program execution. Conducted comprehensive unit and performance testing, including microbenchmarks for SQL injection and cryptographic key leakage scenarios to demonstrate the tool's effectiveness.

Expansion Microscopy

Implemented a 3D convolutional neural network, TrailMap, to segment dopamine axons from light-sheet microscopy volumes. Automated a laborious data pre-processing procedure. Utilized OpenCV to generate scale-invariant feature transform keypoints and estimate a partial 2D affine transformation between pre- and post-expansion images in order to quantify expansion factor. Designed a post-processing pipeline for TrailMap segmentations to extract accurate physical measurements of individual dopamine axon segments.

autocap

Engineered an automated image-captioning application using deep learning for Harvard's Advanced Practical Data Science course. Developed a lightweight Flask API for model inference and a single-page React application. This allowed users to customize model settings and featured an innovative "attention" overlay, providing visual insights into language model predictions. Automated app deployment and Kubernetes cluster creation in GCP using Ansible Playbooks.

autodiff

Developed a Python automatic differentiation library for Harvard's Systems Development course. Constructed a comprehensive pytest testing suite to assess the functionality of core data structures and elementary operations. Created a continuous integration workflow using GitHub Actions and Codecov. Built and distributed package to the Python Package Index.

airshare

Peer-to-peer socket utility that allows clipboard data to be continuously shared between multiple computers. If continuity is enabled on Apple devices, an iPhone's clipboard can be shared to a Windows PC.

Convolutional Neural Networks and Biological Vision

Experimenting with convolutional neural networks to model the development and robustness of biological vision. Analyzed the effects of perturbations on network performance using custom PyTorch models.