Linux systems engineer comfortable from user-space down to the kernel boundary — building and debugging services, squeezing performance out of GPUs, and automating fleets of Linux machines. Contributor to a live 100-node edge Linux deployment.
Projects
Leopard Detection System — Edge Linux Runtime
Built and debugged the Linux runtime powering a 100-node edge deployment.
- Edge runtime as Linux services: packaged the detector as systemd units with auto-restart, resource limits and journald logging on every Jetson.
- GPU optimization: profiled the inference path with Nsight / nvidia-smi, applied TensorRT and memory tuning to raise GPU utilisation from ~50% to >90% and cut per-frame time.
- Low-level debugging: diagnosed crashes and stalls with gdb (core dumps), strace/ltrace (syscalls) and perf (hotspots) — not print debugging.
- Fleet automation: Bash scripts running commands in parallel across 100 Linux nodes over SSH (xargs -P / pssh) for upgrades, health sweeps and log collection.
- Remote debugging & packet analysis: SSH-over-VPN into nodes; tcpdump/Wireshark to root-cause failing API calls at the TCP/TLS level.
Linux Systems Programming
- Processes & threads: pthreads, fork/exec, scheduling, resource limits.
- IPC & sockets: shared memory, pipes, signals, and TCP/UDP socket programming.
- Containers under the hood: Docker via Linux namespaces & cgroups.
Advanced AI Ecosystem Training
by DeepManthan AI Production AI Ecosystem Program
Intensive program on building & operating production AI systems — containers, orchestration, distributed systems, networking and deep learning.
Containerisation — Docker, images, isolation, OTA rollouts
Orchestration — Kubernetes, self-healing, scaling
Distributed Systems — Raft consensus, consistent hashing
Messaging — pub/sub, topics, partitions, offsets
Concurrency — multi-threading, producer–consumer
Computer Vision — CNNs, convolution, detection
Networking — TCP/IP, SSH, private overlays
Linux & Ops — processes, daemons, systemd
Impact Highlights
100
Linux nodes operated
>90%
GPU utilisation after tuning
hrs → sec
fleet-wide ops via parallel SSH
0
reliance on print debugging