Skip to content
Open to full-time roles

Sushank
Gundrathi

Rust Systems Engineer

I build GPU-accelerated video pipelines and low-latency backend systems in Rust — currently turning live camera feeds into real-time crowd-safety intelligence at Fraunhofer IOSB.

Get in touch
Berlin, Germany––:––

the kind of pipeline I ship every week

65%

RPC latency cut

p99 latency reduction via pooled, multiplexed transport

6.5×

Multi-core speedup

Rayon work-stealing parallelism + zero-copy buffer reuse

30

Cameras, one GPU

Live streams time-multiplexed through a single AI inference pipeline

<1s

Glass-to-glass latency

RTSP ingest to browser via WebRTC simulcast delivery

01 · About

From a small town in Telangana,India to the hot path in Berlin

I grew up in a small town in Telangana, India, fought my way through one of the most competitive entrance exams into NIT Surat, and then made a bet on myself: moved to Berlin in 2024 to go all-in on systems engineering.

That bet is paying off. At Peer Network I went from onboarding to owning the entire Solana smart-contract layer of a social-token platform — and when the senior backend engineer fell ill, I picked up the core backend too. Today at Fraunhofer IOSB I'm the engineer behind the Rust backend of a real-time crowd-safety platform: GPU video pipelines, custom GStreamer plugins, AI inference on live streams — software that exists to prevent the next Love Parade disaster.

I like the unforgiving parts of the stack — the hot path where a dropped frame, a lock held too long, or an extra copy is visible to everyone. Rust is my home base, but the constants are lower-level: concurrency, memory, protocols, and profiling before optimizing.

Base
Berlin, Germany
Focus
Real-time streaming & systems
Daily drivers
Rust · C/C++ · GStreamer
Languages
English · Telugu · Hindi · German(A2)

Currently

  • Building GPU-accelerated video pipelines at Fraunhofer IOSB
  • Master's thesis on memory-safe real-time inference pipelines
  • Deepening distributed-systems and system-design depth
02 · Experience

Where I've shipped

Research institutes, a blockchain startup, government-scale infrastructure, and a space agency — always the performance-critical part.

Software Engineer, Video Streaming — Working Student

Fraunhofer IOSB · Video Analysis Systems (VID)

Mar 2026 — Present

Berlin, Germany

Sole owner of the Rust backend of a production crowd-safety video platform — the engine that ingests live RTSP camera feeds, decodes H.264/H.265 on GPU, runs ONNX crowd-density inference on every frame, and serves enriched video back out over RTSP and WebRTC.

  • Design and build custom GStreamer plugins in Rust (gstreamer-rs) — elements, caps negotiation, buffer flow, and custom GstMeta — working directly against GStreamer internals over FFI.
  • Built a round-robin mux/demux plugin pair that time-multiplexes N live cameras through one AI inference pipeline, cutting GPU/model cost from N instances to one; lock-free slot scheduling verified at 30 concurrent cameras.
  • Own the gRPC (tonic) interface between the Rust backend and the TypeScript/Deno middleware, plus dynamic per-camera pipeline construction and teardown at runtime.
  • Ship production-grade code under strict CI: clippy -D warnings, rustfmt, integration tests against mock ONVIF/RTSP servers, GitLab pipelines, Docker with NVIDIA Container Toolkit for GPU passthrough.
  • Profile and optimize the hot path — GPU decode via NVDEC/CUDA, zero-copy buffer strategies, frame-drop and latency budgets at 30 fps per camera.
  • Rust
  • GStreamer
  • CUDA / NVDEC
  • ONNX Runtime
  • WebRTC
  • RTSP
  • gRPC
  • Docker
  • GitLab CI

Rust Backend & Blockchain Engineer — Working Student

Peer Network PSE GmbH

Feb 2025 — Feb 2026

Berlin, Germany

Owned the Solana layer of a social-token platform end to end — token program, tokenomics, and the web2↔web3 bridge — then crossed over into the core backend to fix production bugs across the social graph and payments APIs.

  • Rust
  • Solana
  • Anchor
  • SPL Token 2022
  • Orca Whirlpools
  • Axum
  • Tokio
  • GraphQL
  • PostgreSQL

Software Engineer

Elite Soft · Government Education Board

Sep 2023 — Sep 2024

Hyderabad, India

DevOps and platform work for a government education platform serving tens of thousands of concurrent users.

  • Docker
  • GitLab CI/CD
  • Linux
  • Bash
  • Monitoring

Research & Development Intern

National Remote Sensing Centre — ISRO

Jan 2023 — Jun 2023

Hyderabad, India

Satellite-imagery restoration research at India's national space agency's remote sensing centre.

  • Python
  • TensorFlow
  • OpenCV
  • GANs
03 · Work

Systems I've built

Real production work — GPU video pipelines, on-chain token economies, and the profiling scars to prove it. Open the case notes on any card.

Featured · Fraunhofer IOSB

Real-time crowd-safety video intelligence

The problem

Crowd disasters like the 2010 Love Parade tragedy share one failure mode: human operators cannot watch dozens of camera feeds and spot dangerous density buildup in time. Passive surveillance sees the disaster — it doesn't prevent it.

The system

A production-grade platform that ingests live IP-camera streams, decodes them on GPU, runs crowd-density AI on every frame, and pushes live heatmaps, alerts, and per-m² density metrics to operators — turning surveillance infrastructure into an active safety system for festivals and stadiums.

My part — I own the backend — the performance-critical engine of the platform. Every frame the system analyses flows through code I design, build, profile, and maintain in Rust.

Data path

  1. 01IP cameras (ONVIF) → RTSP ingest → dynamic per-camera GStreamer pipelines built and torn down at runtime
  2. 02H.264/H.265 decoded on GPU (NVDEC/CUDA), frames kept in GPU memory with zero-copy buffer handling
  3. 03Custom Rust GStreamer plugin runs ONNX crowd-density inference in-pipeline; results attached to buffers as custom GstMeta
  4. 04Metadata survives encode/transport via SEI/KLV injection; camera calibration maps detections to real-world geo-coordinates
  5. 05Enriched streams served back out over RTSP and WebRTC (LiveKit, simulcast) with sub-second latency; control plane over gRPC

1 GPU, 30 cameras

Round-robin mux/demux plugins time-multiplex N live streams through a single inference instance instead of N.

No frame left behind

Stale-frame reuse keeps every output live when a camera stalls — no output ever goes black.

Hermetic testing

Mock ONVIF and RTSP servers let the full pipeline run in CI without a single physical camera.

Geo-referenced detections

In-stream calibration metadata maps image coordinates onto real-world maps for operators.

  • Rust
  • GStreamer
  • CUDA / NVDEC
  • ONNX Runtime
  • WebRTC / LiveKit
  • RTSP
  • gRPC
  • ONVIF
  • Docker
Fraunhofer IOSBN× GPU cost → 1×

Round-Robin Video Inference Multiplexer

A pair of custom GStreamer plugins that funnel N live camera streams through one AI inference pipeline and route results back to N outputs.

  • Rust
  • GStreamer
  • ONNX Runtime
  • LiveKit / WebRTC
Fraunhofer IOSBSub-second glass-to-glass

Live Multi-Camera Streaming Backend

Core backend ingesting RTSP feeds from distributed IP cameras into dynamic GStreamer pipelines, streamed to browsers with sub-second latency.

  • Rust
  • GStreamer
  • WebRTC
  • RTSP
  • ONVIF
  • gRPC
Fraunhofer IOSBPixels → map coordinates

In-Stream Metadata & Georeferencing

The metadata layer that carries structured data — camera ID, KLV, SEI, calibration — inside live video streams, surviving encode and transport.

  • Rust
  • GStreamer
  • KLV / SEI
  • Linear Algebra
Peer Network PSE19.5M tokens, one-shot locked mint

Peer Token — SPL Token 2022 Platform Economy

The full on-chain economy of a social platform: token program, fee-splitting transfers, liquidity pools, and a web2↔web3 bridge.

  • Rust
  • Solana
  • Anchor
  • SPL Token 2022
  • Orca Whirlpools
  • TypeScript
Peer Network PSEp99 latency −65%

Low-Latency RPC Middleware

Replaced per-request connections with a pooled, multiplexed transport for real-time data services.

  • Rust
  • Tokio
  • gRPC
  • Axum
Peer Network PSE6.5× throughput

Parallel Processing Engine

CPU-bound workload parallelized with work-stealing scheduling and profiled down to allocation hotspots.

  • Rust
  • Rayon
  • Flamegraph
  • Criterion
ISRO — NRSCMeasured on PSNR / SSIM

Satellite Image Restoration with GANs

GAN-based restoration for satellite imagery — removing cloud and fog from multi-spectral scenes at India's national space agency.

  • Python
  • TensorFlow
  • OpenCV
  • GANs
04 · Skills

Tools of the trade

Rust is home base. The constants underneath are concurrency, memory, protocols — and profiling before optimizing.

Languages

  • Rust
  • C/C++
  • Go
  • Python
  • TypeScript
  • SQL
  • Bash

Streaming & Multimedia

  • GStreamer
  • RTSP / RTP
  • H.264 / H.265
  • NVDEC / CUDA
  • WebRTC
  • ONNX Runtime
  • ONVIF
  • OpenCV

Systems & Concurrency

  • Tokio
  • Rayon
  • Zero-copy design
  • Lock-free structures
  • FFI
  • Profiling / Flamegraph
  • Memory management
  • Linux

Backend & Protocols

  • gRPC / tonic
  • Axum
  • REST
  • GraphQL
  • PostgreSQL
  • TCP/IP
  • Protobuf

Blockchain

  • Solana
  • Anchor
  • SPL Token 2022
  • Orca Whirlpools
  • Tokenomics design
  • PDAs & CPIs

DevOps & Tooling

  • Docker
  • GitLab CI/CD
  • DevContainers
  • NVIDIA Container Toolkit
  • Just
  • Benchmarking
  • Git
05 · Education

Foundations

2024 — 2026

M.Sc. Computer Engineering

University of Europe for Applied Sciences · Berlin, Germany

  • Thesis (in progress): memory-safe systems design for real-time multimedia inference pipelines — GStreamer plugins in Rust decoding live RTSP H.264/H.265 and running per-frame inference at production frame rates, benchmarked against equivalent C pipeline elements.

2020 — 2024

Bachelor of Technology

National Institute of Technology (NIT), Surat · Surat, India

  • Admitted through JEE Mains — ranked in the top 1% of 1.5 million candidates.
  • 500+ data-structures & algorithms problems solved; active in hackathons.
06 · Contact

Let's build something fast

Hiring for systems, streaming, or backend work — or just want to talk shop about GStreamer and Rust? My inbox is open.

Berlin, Germany––:––

Open to full-time roles — happy to share a detailed CV on request.