Rust C Go SystemVerilog Linux Kotlin/JNI
Building from the kernel up in Bengaluru, India.
struct Engineer {
focus: [&'static str; 3],
rule: &'static str,
}
const MANNAN: Engineer = Engineer {
focus: ["systems", "networks", "hardware"],
rule: "understand the layer below",
};“Here's to the crazy ones. The misfits. The rebels.”
I build close to the machine: network paths, observability pipelines, Android internals, and RTL. I like finding the hidden bottleneck, understanding the boundary, and replacing hand-waving with something measurable.
My philosophy: “Do not cry over spilt milk.” Iterate quickly, fix what broke, keep moving.
Away from the keyboard: technical analysis, science documentaries, and pulling films apart frame by frame.
| Project | What it does | Under the hood |
|---|---|---|
| Punch | Temporary, direct device-to-device connections without accounts or a persistent VPN. Ships file transfer, port forwarding, piping, and an embeddable sidecar. | Rust · Go · QUIC · X25519 · ChaCha20 |
| Sennet | Self-hosted connected observability across traces, logs, metrics, agents, networks, and financial workflows. Keeps correlation, timing, and causality boundaries explicit. | Go · Kafka · ClickHouse · PostgreSQL · OpenTelemetry |
| Termi | An Android terminal with a Rust PTY core and a native Compose interface. The shell path works on-device; SAF-backed I/O and packaging are still being built. | Rust · Kotlin · JNI · PTY · Android SAF |
| Heterogeneous Spatial Tensor Cluster | RTL building blocks for sparse, weight-stationary tensor compute: MAC and PE control, gated execution, double-buffered SRAM, and AXI4-Stream ingestion. | SystemVerilog · MAC/PE · Clock gating · AXI4-Stream · SRAM |
More things I have built
systems programming ████████████████████
networking ██████████████████░░
observability ██████████████████░░
computer architecture ████████████████░░░░
useful interfaces ██████████████░░░░░░
I care about software that owns its data path, states its limits, and leaves enough evidence to explain what happened.
If something here helps you build, debug, or understand the layer below, you can support the next experiment:
make it work → measure it → make it honest → make it fast