Srihari Kompella
I'm studying Computer Engineering at UIUC, graduating May 2027.
My academic interests include Autonomous Systems, Generative Models, and Reinforcement Learning. I've worked on RF signal classification, task-oriented communications, and RL for target tracking; lately I've been building distributed systems and autonomous agents.
Experience
- Nexcepta · a hierarchical RL agent that detects and neutralizes OS- and network-level threats in real time, plus out-of-band telemetry for a Xen/DRAKVUF malware sandbox. Earlier: RF data generation with VQ-VAEs for signal classifiers. 2024, 2026–present.
- U.S. Army Research Laboratory · deployed the Track-MDP algorithm as a real-time gRPC pipeline for autonomous target tracking; built Gymnasium environments and moved training from A2C to PPO. 2025–26.
- U.S. Naval Research Laboratory · a task-oriented communications system that sends images over channels as noisy as −20 dB SNR with a variational information bottleneck. 2025.
- Vedavaag Systems · led and shipped a Flutter rural-banking app used by 4,400+ State Bank of India employees. 2024.
Projects
- distqueue · a fault-tolerant distributed task queue with Raft written from scratch, verified linearizable with Porcupine. Go.
- Time-aware federated multimodal perception · asynchronous LiDAR + camera fusion across separate autonomous platforms for 3D semantic segmentation, without sharing raw sensor data. PyTorch.
- RISC-V kernel · a Unix-like kernel with Sv39 virtual memory, preemptive multitasking, a custom file system, and a shell with fork/exec. C.
- trackmdp-iobt · Track-MDP for testing at the IoBT-MAX facility. Python.
- RT-object_detection · real-time object detection with CPU multiplexing.
- herdr (my fork) · my experiments on top of herdrdev/herdr, the runtime coding agents live on. Rust.