Anirudh Vemuri
Computer Science Student & Software Engineer
I'm a Computer Science student at the University of Michigan with a passion for building high-performance systems and intelligent applications. My work spans from low-level systems programming to machine learning and full-stack development.
Currently exploring the intersection of AI and systems engineering, with experience in microservices architecture, real-time data processing, and end-to-end encryption.
Education
GPA: 3.8
Relevant Coursework
Data Structures and Algorithms, Machine Learning, Web Systems, Computer Organization, Object-Oriented Programming, Discrete Math, Probability Theory, Linear Algebra
Activities
Honors Program, Michigan Hackers Advanced ML Team, Michigan Data Science Team
Experience
AWS Coral
Data Infrastructure + Applied ML
Algorithm Engineering
Machine Learning Research
Projects
- •Designed and implemented a cross-platform P2P file transfer system supporting 100+ concurrent peers with <5 ms peer discovery latency, leveraging UDP-broadcast discovery and multi-threaded connection management
- •Encrypted 500k+ chunks end-to-end using X25519 and ChaCha20-Poly1305, maintaining <0.5 ms latency
- •Achieved 150 MB/s LAN throughput with ZSTD-compressed chunked I/O, reducing payloads by 60%
- •Enabled <100 ms transfer recovery after network interruptions through per-chunk state persistence
- •Reduced the compute cost of LLM calls by up to 30% through developing an algorithm to simplify complex queries with a multi-stage NLP-based query parser leveraging spaCy dependency parses and semantic role labeling
- •Designed an energy-benchmarking dashboard to visualize lifetime and session-level savings with various metrics
- •Implemented a token-based game module to reward calculated energy savings and incentivize user retention
- •Delivered control of 20+ desktop actions with hand gestures with 90% accuracy through CV gesture classification
- •Integrated ASL-based text input through utilizing MediaPipe Hands tracking for <75 ms frame translation latency
- •Improved text input speed by 35% on average by integrating a seq2seq LSTM autocomplete module
- •Minimized unknown sign errors for non-standard ASL words by implementing a k-NN–based semantic map