Skip to main content

B.Tech AI at Kathmandu University · MLOps & AI/ML Engineer

Safal Kumar
Shrestha

AI Developer and Systems Engineer designing real-world technology footprints in Nepal. Specialized in custom production apps under bala.ai, heavy cloud infrastructure virtualization, Kubernetes clusters, and automated smart edge intelligence systems.

KU Krishi App — ICT Award Semi-Finalist
Hydroponics AI System — STEAM Grant Funded
Bare-Metal Infrastructure — Self-Hosted Public Server
AIHPC Group — Kubernetes & HAProxy Cluster
MLOps / Kubernetes Flutter SDK HAProxy / Infrastructure Edge Computer Vision Nepal Systems Engineer 🇳🇵
View Selected Case Studies ↓
Professional Background 01 / 09

Chitwan, Nepal — 2026 Portfolio

Institution Kathmandu University (KU)
Systems Core Kubernetes, HAProxy, Linux Clustering
Google Developer Name bala.ai · 3 Live Deployments
Cloud Setup Self-Built, Publicly Hosted Node
National Honor KU Krishi — Selected for ICT Award
Research Funding Hydroponics AI — STEAM Grant Recipient

Bridging the gap between raw machine learning models and hard bare-metal infrastructure orchestration.

I'm Safal Kumar Shrestha, an Artificial Intelligence developer and Systems Engineer pursuing a B.Tech in AI at Kathmandu University. My engineering methodology looks past simple software development; I design, build, and maintain the underlying infrastructure required to run high-performance algorithms in production.

Whether deploying decentralized mobile platforms on Google Play via bala.ai, managing custom container orchestration routines, or configuring high-availability network routers, I build for scale. My foundational systems journey includes building a completely independent, self-hosted bare-metal server network from scratch to power live application backends, alongside complex core architecture research inside Kathmandu University's Department of AI.

I'm an active member of the KU Artificial Intelligence Club (KUAIC), where I collaborate with peers on AI research, workshops, and community projects that bridge the gap between academic AI and real-world deployment in Nepal.

Engineering Case Studies

Selected Technical
Deployments.

02 / 09
01

KU Krishi App

National ICT Award Google Play Store Production
Source Code

An embedded AI agricultural assistant built at Kathmandu University to resolve crop threats on localized farms. It uses edge computer vision algorithms to instantly isolate agricultural diseases like late blight through smartphone optics, outputting adaptive metrics natively in the Nepali language. Selected as a semi-finalist for Nepal's national ICT Award.

Flutter SDK Python Engine TensorFlow Mobile YOLOv8 Weights FastAPI Server Firebase Cloud OpenCV Pipeline
02

Hydroponics AI System

STEAM Research Grant DFRobot Hardware Sponsored
Hardware Schema

An industrial IoT automation platform applying Reinforcement Learning (RL) frameworks to soil-free smart urban farming. Funded by the Ministry of Education, Science and Technology under the STEAM Grant with hardware telemetry equipment sponsored directly by global robotics manufacturer DFRobot.

Embedded C/C++ TensorFlow Core PyTorch RL OpenCV Matrix ESP32 SoC Raspberry Pi Arduino Drivers MQTT Protocol WebSockets FastAPI Backend Flutter Dashboard Gymnasium LiveKit Video Stream

🔌 How DFRobot Became Our Hardware Partner

I took the initiative to reach out directly to DFRobot via email, presenting our research vision for the Hydroponics AI system and explaining how their precision industrial instrumentation could accelerate our work. After reviewing our project proposal and STEAM grant selection, DFRobot officially came on board as a hardware sponsor — gifting us essential electronic components including pH probes, Co2, DHT22, Spectrometer Sensor, ESP32-S3 AI Camera, ESP32-P4 and WaterProof Ultrasonic sensor. Their support transformed our sensor array from theoretical designs into a fully functional, real-time edge processing data pipeline.

What I Engineered

  • Programmed the embedded control layer on ESP32 and Raspberry Pi to interface with pH, EC, temperature, and humidity sensors — using DFRobot's gifted hardware components for the sensor array
  • Implemented sensor data acquisition and actuator control to manage pumps and nutrient dosing mechanisms via MQTT
  • Developed the backend infrastructure for the computer vision pipeline, enabling image uploads and AI model inference
  • Developed a FastAPI backend with MQTT and WebSocket support for real-time device monitoring and automation
  • Integrated LiveKit for real-time video streaming to enable live monitoring of plant growth
  • Built a Flutter mobile application for monitoring sensor data, live streams, and remote control
  • Deployed the system on a local edge server to support real-time processing and device communication
  • The project proposal was selected and funded by STEAM, with hardware components gifted by DFRobot to support research in AI-powered sustainable agriculture
03

Self-Hosted Public Cloud Infrastructure

Bare-Metal Hardware Stack Semester Project — Tiffin Backend

How do you avoid expensive public cloud bills while keeping full low-level systems control? In 2022, I architected, built, and provisioned an independent bare-metal server node from scratch and configured it to be publicly accessible from the open internet. This custom hardware rig hosted the production data endpoints, security rules, and real-time transaction pipelines for the Like Tiffin mobile application ecosystem — proving that a student-built infrastructure could reliably serve real users without relying on AWS, GCP, or Azure.

Bare-Metal Linux Architecture Public IP Networking Reverse Proxy Routing Secure SSH Layer Node.js Daemon Processing Hardware System Maintenance

Infrastructure Engineering Steps

  • Assembled the standalone server hardware, managing cooling limits and computing power requirements
  • Configured core Linux network settings, static IP routers, and network firewalls to allow incoming traffic safely
  • Set up reverse proxy configurations to manage domain handling and enforce SSL encryption layers
  • Maintained perfect database and API uptime on raw physical hardware to reliably serve mobile clients under pressure
04

AIHPC — High-Performance Computing Cluster

KU Department of AI

In 2025, I led an intensive system orchestration initiative within Kathmandu University's Department of AI to build a dedicated high-performance computing cluster (AIHPC). The project involved configuring Kubernetes containers for scaling distributed worker loads and establishing HAProxy layers to direct load balancing models across system networks.

Kubernetes (K8s) HAProxy Load Balancing NVIDIA RTX 3080Ti Docker Containers Linux Administration GPU Cluster Topology

Engineering Post-Mortem: Why It Failed

  • The hardware environment was limited because consumer-grade NVIDIA GeForce RTX 3080Ti GPUs lack hardware-level architecture partitioning (Multi-Instance GPU / MIG)
  • This constraint prevented stable multi-tenant GPU virtualization across the cluster
  • Localized technical team knowledge scaling boundaries (skill issues) further compounded the hardware limitations
  • The project provided deep insight into the practical boundaries between enterprise grid computing systems and consumer accelerator units
05

TrekSathi

Google Play Store Production

A high-reliability production trekking companion app engineered to optimize safety workflows for tourists tracking remote mountain corridors across Nepal. Features custom vector trail tracking layers, elevation mapping charts, automated atmospheric data caching, and offline GIS coordinates lookup profiles. Published on Google Play under bala.ai.

Flutter Frontend Dart Google Maps SDK Django REST API MongoDB Server OpenWeather API Integration Firebase Cloud

What I Engineered

  • Built the complete Flutter app from scratch — UI design, state management, navigation, and all screens
  • Integrated Google Maps SDK with custom trail overlays, waypoint markers, and elevation profile charts
  • Collected and structured raw trekking trail data into a reliable geospatial dataset
  • Connected OpenWeatherMap API for hyper-local mountain weather forecasts with severe weather alerts
  • Developed a Django + MongoDB backend with REST APIs for trail data, trip history, and trekker profiles
  • Implemented offline map caching for use without internet in remote mountain areas
  • Added agentic AI features for autonomous trek planning suggestions and personalized route recommendations
  • Managed the full Play Store release pipeline under the bala.ai account
06

Like Tiffin App

Google Play Store Production
View Deploy Client

A home-cooked meal delivery platform connecting students and office workers with local home chefs, hotels, and restaurants in Nepal. Like Tiffin promotes healthy eating, reduces food waste, and gives home cooks an income stream — all through a clean, fast Flutter app published under bala.ai. Each restaurant or hotel has its own cloud kitchen, allowing customers to pre-order meals before arrival. Powered by my self-hosted bare-metal server infrastructure.

Flutter Native Dart UI Engine Firebase Storage Firestore Sync Streams Node.js Engine Google Maps SDK

What I Engineered

  • Designed and developed the full Flutter cross-platform app for customers and home chefs
  • Engineered instant real-time order stream tracking using Firebase Firestore listeners
  • Implemented Google Maps-based delivery radius matching customers to nearby chefs
  • Created a centralized Node.js admin backend with APIs for user management, order history, and analytics
  • Handled full Play Store publishing under bala.ai

Technical Capabilities Matrix

Core Tooling &
Technologies.

03 / 09
MLOps & Core Systems
Kubernetes (K8s) HAProxy Load Balancing Bare-Metal Server Architecture Linux Systems Administration Reverse Proxies & Network Gateways GPU Cluster Resource Profiling
AI / Machine Learning
TensorFlow PyTorch Architecture YOLOv8 Object Detection OpenCV Processing Scikit-learn Computer Vision Models Natural Language Processing
Mobile Systems
Flutter SDK Engine Dart Native Android Systems (Java) Firebase Cloud / Firestore Google Play Console Publishing REST API Integration
Backend Engineering
Python FastAPI Framework Django REST Node.js Engine MongoDB MySQL Infrastructure
IoT & Hardware
Raspberry Pi Systems Arduino Frameworks ESP32 Microcontrollers MQTT Protocols Sensor Integration DFRobot Hardware SDK
Engineering Tools
Git & GitHub Docker Containers CI / CD Pipelines Linux Bash Agile / Scrum

Professional Chronology

Work History &
Research Post-Mortems.

04 / 09

STEAM Research Grant

STEAM Funding Council — Ministry of Education, Science and Technology, Government of Nepal

Awarded competitive scientific research capital for validating Hydroponics AI systems via Reinforcement Learning algorithms.

STEAM Research Grant — Officially Funded

DFRobot Hardware Sponsorship

DFRobot — Global Robotics & Open-Source Hardware Provider

Initiated direct outreach via email to DFRobot, presenting the Hydroponics AI research vision. Successfully secured hardware sponsorship — DFRobot gifted precision industrial instrumentation modules (pH, Co2, ESP32-S3 Cam) to support the sensor infrastructure for the smart farming system.

DFRobot — Hardware Gifted

High-Performance Computing Core Engineer (AIHPC Initiative)

Department of Artificial Intelligence — Kathmandu University

Led an intensive system orchestration initiative to build a dedicated high-performance computing cluster (AIHPC) within the university's AI department. Configured Kubernetes containers for scaling distributed worker loads and established HAProxy layers to direct load balancing models across system networks.

🔍 Engineering Post-Mortem: The hardware environment was limited because consumer-grade NVIDIA GeForce RTX 3080Ti GPUs lack hardware-level architecture partitioning (Multi-Instance GPU / MIG). This constraint, paired with localized technical team knowledge scaling boundaries (skill issues), prevented stable multi-tenant GPU virtualization. The project provided deep insight into the practical boundaries between enterprise grid computing systems and consumer accelerator units.

Kubernetes Cluster Pods HAProxy Reverse Routing NVIDIA Driver Topology Hardware Constraint Diagnostics

ICT Award — Semi-Finalist

National ICT Awards — Nepal

KU Krishi was selected as a semi-finalist for Nepal's national ICT Award for its innovative use of computer vision and AI to empower smallholder farmers.

ICT Award — Semi-Finalist

Software Engineer, Android (Full-Time Remote)

Shop2App Corp — United States (Remote)

Engineered, deployed, and managed enterprise-tier native mobile commerce layouts connecting native Android systems with the global Shopify core data ecosystem. Worked across the full development lifecycle from feature design to Play Store release, enabling merchants to run their Shopify stores as branded Android apps.

Java Android SDK Shopify API REST Webhooks Google Play Console

Feb 2022 – Jul 2022 · 6 months · United States (Remote)

Academic Qualifications

Formal Engineering
Foundations.

05 / 09
Kathmandu University (KU)
Dhulikhel, Kavrepalanchok · Nepal · Est. 1991
B.Tech in Artificial Intelligence (Bachelor of Technology)
Department of Artifical Intelligence | School of Engineering
Active Enrollment

Acquiring deep structural expertise at Nepal's premier academic research institution. Core syllabus focus balances advanced matrix mathematics, applied data structures, computer system architectures, deep neural optimization, and digital system frameworks designed to address modern software challenges.

Key Modules

Machine Learning Deep Learning Computer Vision Natural Language Processing Data Structures & Algorithms Probability & Statistics Linear Algebra Database Systems Software Engineering Research Methodology
Enrolled
Expected
Degree
B.Tech
Major
Artificial Intelligence
ICT Award
KU Krishi
STEAM Grant
Hydroponics AI
Nepal Polytechnic Institute
Bharatpur · Nepal
Diploma in Computer Engineering
Graduated

Studied at one of Nepal's well-regarded technical institutes, with a curriculum covering programming, electronics, networking, data mining, and database management.

Key Modules

Mobile Development Web Development C/C++ Git/GitHub Networking Database Management
Enrolled
End
Degree
Diploma
Major
Computer Engineering

Community

KU Artificial Intelligence Club

KUAIC — Kathmandu University

2022 – Present · Active Member

Hackathon

AI Crusade

KUAIC 2024

2024 · 3rd Place

Award

Nepal ICT Award

National ICT Awards · KU Krishi

2024 · Semi-Finalist

Grant

STEAM Research Grant

STEAM Programme · Hydroponics AI

2026 · Funded

Beyond the Screen

On the
trail.

06 / 09

When I'm not writing code, I'm out in the mountains. Trekking is where I disconnect and reset — no IDE, no stack traces, just altitude, trail and silence. Nepal's trails are part of who I am, and they're part of why I built TrekSathi.

Annapurna Base CampNorth ABC Trek
4,250 m peak elevation Annapurna region

My first trek along the recently popular North ABC route (2025/26), following the winding path of the Modi Khola river up to Annapurna Base Camp (ABC) at 4,130 m.

Gandaki Province 4,250 m
Mardi HimalViewpoint Trek
Approx. 4,500 m highest point Annapurna region

A shorter Mardi Himal trek to the viewpoint, offering a ridge walk above the treeline. We enjoyed every moment of the journey with friends and fellow trekkers along the trail.

Highlights

  • Walked the exposed high ridge between Low Camp, High Camp, and the viewpoint
  • Enjoyed the camaraderie with friends and fellow trekkers
  • Experienced long stretches of trail in relative solitude
  • Descended via the Forest Camp route through dense bamboo and rhododendron jungle
Kaski District · Gandaki Province Approx. 4,500 m

Common Questions

Frequently
Asked.

07 / 09

Who is Safal Kumar Shrestha?

Safal Kumar Shrestha is an AI Developer, MLOps Engineer, and Android Software Engineer from Chitwan, Nepal. He is currently pursuing a B.Tech in Artificial Intelligence at Kathmandu University and publishes apps under the name bala.ai on Google Play. His work has been recognized with the Nepal ICT Award (Semi‑Finalist) and a STEAM research grant.

What are Safal's most notable projects?

Key projects include KU Krishi (ICT Award Semi‑Finalist, computer vision for crop disease detection), Hydroponics AI (STEAM Grant & DFRobot sponsored, AI + IoT automated farming), a self‑hosted bare‑metal server powering the Like Tiffin app, an AIHPC Kubernetes cluster, and the TrekSathi trekking companion.

What technologies does Safal specialize in?

Safal works across the full stack: MLOps (Kubernetes, HAProxy, Docker), AI/ML (TensorFlow, PyTorch, YOLOv8, OpenCV), mobile (Flutter, Android/Java), backend (FastAPI, Django, Node.js, MongoDB), and IoT hardware (ESP32, Raspberry Pi, Arduino, DFRobot sensors).

How can I collaborate with Safal?

You can email him at dev.safalstha@gmail.com or connect on LinkedIn. He is open to research collaborations, AI/ML internships, app development partnerships, and remote roles.

Does Safal have experience with cloud infrastructure?

Yes. He built a completely self‑hosted bare‑metal server that is publicly accessible and powered the Like Tiffin app backend. He also led the AIHPC initiative at KU, configuring Kubernetes and HAProxy on NVIDIA RTX 3080Ti GPUs, gaining deep insights into enterprise‑grade cluster orchestration.

Gallery

Captured
moments.

08 / 09
Loading gallery...
Contact Available for opportunities

Let's build
something.

08 — 08

Open to research collaborations, AI/ML internships, app development partnerships, or a genuine conversation about building things for Nepal. Based in Kathmandu — remote-friendly.

Email

dev.safalstha@gmail.com

Find me online

LinkedIn
in/itsmesafal ↗
GitHub
Source code ↗
Google Play — bala.ai
3 published apps ↗
Chitwan, Nepal