Samir Sen

Kathmandu · --:-- NPT · Open to collaborations

Systems that see, think and move.

R&D AI roboticist and automation engineer at Baliyo Ventures. I turn computer-vision research into reliable, physical-world automation, one perception loop at a time.

Selected work Download CV
  • 01 Perception OpenCV · PyTorch · SSD
  • 02 Edge compute Jetson Nano · Lidar
  • 03 Actuation Embedded control

27.7172° N  85.3240° E Portfolio — Vol. 03 / 2026

01 Profile (About)

Curiosity, made tangible.

Fig. 01 — Perception head Live render

I work at the intersection of intelligent software and the physical world — building systems that are perceptive, connected, and ready for action beyond the notebook.

As a junior R&D AI roboticist at Baliyo Ventures, I explore computer vision, edge devices, and spatial intelligence. BSc Hons in Computing with AI, Islington College (2025), plus internships at Ants Pvt. Ltd. and Atreya Software Solutions.

build with intent. learn in public. ship what matters.

  • 06 Selected projects
  • 2025 BSc (Hons) Computing with AI
  • KTM Based in Kathmandu, Nepal

02 Selected work (06 projects)

From signal to solution.

Perception, edge compute and embedded systems — each project starts with a physical-world constraint and ends with something that runs.

CLASS · M · BRG CLASS · M · BRG

Live · SSD MobileNetV2 · FastAPI stream

Fig. 02 — Detection overlay on live feed 01 / 06

01 — Computer vision

Real-time object detection, distance & direction

Turns live video feeds into spatial awareness, detecting objects while estimating their distance and direction for downstream robotic decisions.

Bridges the gap between raw camera input and actionable perception through a lightweight, accessible API.

Model
SSD MobileNetV2
Serve
FastAPI · live video
Output
Class · metres · bearing

02 — Spatial edge AI

Orientation & spatial edge intelligence

Gives an edge device environmental context by combining Lidar and depth sensors with PyTorch inference on a Jetson Nano.

Moves spatial reasoning closer to the sensor, reducing dependence on remote compute for responsive robotics.

Device
Jetson Nano
Sensors
Lidar · depth
Inference
PyTorch, on-device
LIDAR RETURNS + ORIENTATION
Fig. 03 — Lidar returns + orientation 02 / 06

Interlude — Fig. 04 Gait study

Locomotion is a perception problem first.

03 Services & stack (What I do)

A practical stack.

I take a problem from sensor to service: perceive it, model it, ship it to hardware, and expose it through an API someone else can use.

  • (A)

    Perception & computer vision

    Detection, depth and spatial reasoning that turns pixels into decisions a machine can act on.

    • PyTorch
    • OpenCV
    • SSD MobileNetV2
    • Lidar fusion

  • (B)

    Edge & embedded systems

    Models squeezed onto real hardware, where latency, power and memory shape the design.

    • Jetson Nano
    • Microcontrollers
    • Linux · SSH
    • Edge integration

  • (C)

    Backend & architecture

    Services that expose robotic capability cleanly — typed APIs, containers, and sane boundaries.

    • Python · FastAPI
    • Java · Spring Boot
    • REST · Microservices
    • Docker · AWS · Azure

  • (D)

    Data & analytics

    Turning raw logs and datasets into models and charts that hold up to questions.

    • scikit-learn
    • Pandas · NumPy
    • Regression
    • Matplotlib · Seaborn

04 Experience (Path)

Learning by shipping.

  1. 2025 — Now

    R&D AI Roboticist

    Baliyo Ventures

    Exploring computer vision, edge devices, spatial intelligence, and practical robotic systems.

  2. 2025

    BSc (Hons) Computing with AI

    Islington College

    Built a foundation across artificial intelligence, software engineering, data, and systems.

  3. Earlier

    Engineering internships

    Ants Pvt. Ltd. · Atreya Software Solutions

    Learned to move from coursework into team workflows, implementation, and delivery.

Also: Club Tamer for Leo Club of Temple Peace (2021–2022) — where I learned to turn shared energy into organized action.

05 Field notes (Writing)

Work in progress.

Note / 001 Draft

Perception is only useful when it changes a decision.

The next layer after detection is context: what should the system do with what it sees?

Note / 002 Draft

Edge systems make constraints visible.

Latency, power, memory, and deployment shape the model as much as the dataset does.

Note / 003 Draft

Build the smallest loop that can teach you something.

Hardware and software become clearer when the first experiment is close to reality.

06 Contact (Let's build)

Have a difficult problem?

Let’s talk about what you’re building, what’s stuck, and where intelligent systems could take it next.

Write to me

Start a conversation Download CV
GitHub
NotAScratch
LinkedIn
in/samirsen8
Location
Kathmandu, NP
Local time
UTC +5:45
CASE STUDY / 01

PERCEPTION PIPELINE

Real-time object detection

THE PROBLEM

THE APPROACH

THE LESSON