I love to build machines
that see, think, and move

Namaste,
I'm Shiven.

VP Engineering @ CommBox.tv, where I started as the founding engineer.

github / linkedin / x / resume

Perception to action

I work where models meet motors - fine-tuning vision-language-action models, building physics-true simulations, and shipping the unglamorous systems code that makes robots actually work. From CUDA kernels to control loops, I care about the whole stack.

Based
Rohtak, HR, India
Currently
V.P. Engineering @ CommBox.tv
Open to
Contract & Projects Collaboration

19

Projects shipped

11

Certifications

2×

E-Yantra team lead

3×

SIH team lead

Where I've shipped

CommBox

A modern AI-driven cricket media platform.

Type
Full-time
Based
Gurugram, India

Commbox Social Media Development and Management LLC

  1. Oct 2026 - Present

    VP Engineering

    Tech Lead

    Promoted from Founding Engineer to lead all of engineering. I own the technical roadmap, system architecture, and delivery across infrastructure, backend, mobile, and the AI features on the roadmap.

    • Shaping the engineering team through technical hiring, and setting the overall tech vision and direction for the company
  2. Promoted · Oct 2026

  3. Jan 2026 - Oct 2026

    Founding Engineer

    Designed the entire system from scratch: the architecture, cloud infrastructure, media and AI pipelines, and the apps on top of them. Then built and shipped it.

    • about 70%

      Deploy lead time cut

      Production cloud on AWS EKS

      Terraform and Kubernetes infrastructure with GitOps delivery through ArgoCD across 5+ microservice repos, shipped as zero-downtime rolling releases.

    • under 90s

      Median ingest to playback

      Adaptive-bitrate video pipeline

      Multi-rendition HLS and DASH transcoding for 1080p and 4K ingest, at ~35% lower cost per asset.

    • 90%+

      Caption accuracy

      Production auto-captioning

      WhisperX large-v3 with pyannote speaker diarization and LLM post-processing on English and Hinglish audio, writing SRT ~4× faster than real time.

    • about 60%

      Less manual editorial work

      AI video analysis service

      Automated keyframe extraction for thumbnails, scene segmentation, and content tagging.

    • about 50%

      Lower incident MTTR

      End-to-end observability

      Grafana and Prometheus dashboards instrumented across every service.

    • 3

      Platforms, one codebase

      Mobile app and creator CMS

      React Native for iOS and Android plus a creator and admin web CMS, kept in sync in real time over WebSocket and Redis Pub/Sub.

Robora

A web3 robotics startup.

Type
Contract
Based
Remote
Status
Wound down, 2025

Shenzhen Robora Automatic Robots Ltd Enterprise

Robora has since wound down and robora.xyz is offline. Engagement details are available on request.

Robotics Simulation & ML Engineer

Sep 2025 - Dec 2025

Led simulation and ML infrastructure - the VLA SDK, the physics environments, the 3D reconstruction pipeline, and the data tooling around them.

  • Architected a VLA SDK for SmolVLA, π₀ / π₀.5, and GR00T N1.5 - inference plus PEFT fine-tuning (LoRA, QLoRA)
  • Built multi-physics sim environments (PyBullet, MuJoCo, Gazebo) with PID, LQR, MPC, and RL control via SB3 / TorchRL
  • Developed a 3D reconstruction toolkit - COLMAP, GLOMAP, 3D Gaussian Splatting - for ultra-realistic environment modeling
  • Trained a YOLOv11 GDPR anonymization pipeline hitting 95%+ accuracy on faces and license plates

StackPython · PyTorch · MuJoCo · PyBullet · Gazebo · ROS2 · TorchRL · COLMAP · Gaussian Splatting · YOLOv11

Uniconverge

Type
Internship
Based
Noida, India

Uniconverge Technologies Pvt. Ltd.

AI/ML Software Engineer Intern

Jul 2025 - Sep 2025

IP-camera management in modern C++ - RTSP streams forwarded, monitored, and analysed from a native Windows app.

  • Built a native Windows app in C++ and Qt 6.5 for managing LAN IP cameras with RTSP streaming
  • Integrated AWS SageMaker AI for real-time productivity monitoring from camera streams
  • Set up WireGuard VPN tunneling with TCP port forwarding for secure peer-to-peer links to AWS EC2
  • Designed the system-level pieces: system tray integration and a background service

StackC++17 · Qt 6 · RTSP · WireGuard · AWS SageMaker · AWS EC2 · CMake

Independent

Type
Freelance
Based
Remote

Freelance Developer

Feb 2025 - Present

Full-stack web and native Android delivery for clients - scoped, built, deployed, and maintained end to end.

  • Shipping full-stack web apps on Next.js, TypeScript, and Tailwind
  • Building native Android clients with Kotlin and Jetpack Compose
  • Owning cloud deploys and infrastructure on AWS and Vercel

StackNext.js · TypeScript · Kotlin · Jetpack Compose · Spring Boot · AWS

Flagship builds

Desky preview

Desky

A desk robot whose whole brain is the phone you dock into it.

KotlinJetpack ComposeLLM / STT / TTSMQTTEmbedded C
2025 · roboticsLive
Kalakari preview

Kalakari

Doodle for 30 seconds. A CNN in your browser names it.

PyTorchONNX RuntimeNext.jsCNNQuick Draw
2026 · aiGitHub Live
SonyMDR Open preview

SonyMDR Open

Sony's Android-only headphone app, reverse engineered for Linux.

Reverse EngineeringPythonBlueZRFCOMMJADX
2026 · linuxGitHub
VLA Stack preview

VLA Stack

One SDK to fine-tune and deploy five VLA model families.

PyTorchSmolVLAπ₀ / π₀.5OpenVLAGR00T N1.5
2025 · roboticsGitHub
Pithuu OS preview

Pithuu OS

A Linux distro with its own package repo - built and maintained solo.

Arch LinuxLinux KernelGRUBPKGBUILDGNOME
2024 · linuxGitHub Live

Capabilities

01Robotics

Simulation-first robotics: VLA models, classical and RL control, and vision pipelines that survive contact with hardware.

ROS2 · PyBullet · MuJoCo · Gazebo · OpenCV · CoppeliaSim · MATLAB · URDF · COLMAP · Gaussian Splatting

5 projects →
02GPU & CUDA

Hand-written kernels, parallel pipelines, and real-time graphics - performance measured, not guessed.

CUDA · C++ · OpenGL · CMake · Parallel Computing · Ray Tracing

2 projects →
03AI Engineering

From fine-tuning to inference infrastructure: RAG systems, local LLMs, and the ops that keep them serving.

PyTorch · Transformers · LangChain · vLLM · llama.cpp · Ollama · Hugging Face · PEFT

3 projects →
04Android

Native Android with modern everything - Compose, Material 3, and architecture that scales past the demo.

Kotlin · Jetpack Compose · Material 3 · Hilt · Room · Ktor

2 projects →
05Linux Systems

Distro-maintainer-level Linux: kernels, packaging, hardening, and the tooling developers actually want.

Arch Linux · Kernel · Bash · PKGBUILD · systemd · Security

3 projects →
06Full-Stack

End-to-end web applications tuned for real traffic - SSR, websockets, and cloud deploys that hold up.

Next.js · TypeScript · Node.js · Spring Boot · AWS EC2 · Tailwind CSS

3 projects →
Currently exploring - MuJoCo & PyBulletCurrently exploring - Onshape CADCurrently exploring - Multimodal LLMs (Gemma 3n)Currently exploring - TorchRLCurrently exploring - MuJoCo & PyBulletCurrently exploring - Onshape CADCurrently exploring - Multimodal LLMs (Gemma 3n)Currently exploring - TorchRL

Proof of work

Hover to preview - click to verify

Field notes