Open to new opportunities · India, US, EU

Hey there, I’m

Mubashshir Uddin. Deep Learning Engineer, Gurgaon, India

I build

Deep learning engineer at TerraNxt, Gurgaon. Physics nerd by training, builder by habit, from Agra originally. This is my little corner of the internet.

60-second summarySee my workSay hi

Move to pull the stars. Click to blast them away.

The 60-second version.

For recruiters and hiring managers who’d rather skim. Pick what you’re hiring for and I’ll tell you why I fit.

Role
Deep Learning Engineer at TerraNxt (Nov 2025 to now). Before that, Smarttrak AI.
Where
Gurgaon, India. Open to relocating to the US or EU.
Best at
Taking a physics-grounded AI idea all the way to a deployed, monitored system.
Core stack
Python, PyTorch, CUDA, LangGraph, FastAPI, React, Kubernetes, InfluxDB
Education
IIT Kanpur. B.S. Physics, B.Tech Aerospace (second major).
Reach me
Email · LinkedIn ·

I like understanding how things work, then making them work better.

I did Physics and Aerospace Engineering at IIT Kanpur, which mostly taught me one thing: write down the physics first, then let the AI fill in what the equations can’t. That’s still how I work.

These days I spend my time on solar plants. I built the monitoring backbone that watches dozens of sites live, the simulations that tell us how a plant should be performing, and the 3D and VR twins that let you walk around one without leaving your desk.

Outside of work I fall down rabbit holes about relativity, information theory and where AGI might be heading. Agra is home, Gurgaon is where I build, and I’m open to moving to the US or EU for the right adventure.

24solar sites monitored live
43,200data points per second ingested
1,776buildings in my labelled dataset
90%functional accuracy, rooftop footprints and heights

The path so far.

Tap a card to see what I actually did there.

Nov 2025 to now · Gurgaon

Deep Learning Engineer, TerraNxt

Building ScadaNXT, an AI-powered SCADA and operational intelligence platform for utility-scale solar, mostly on my own.

Rooftops from space
  • Built a 3D small-object detection pipeline for rooftops from 30 cm stereo satellite imagery, reaching 90% functional accuracy on building footprints and heights.
  • Labelled a 1,776-building GIS dataset, trained to convergence, and set up a self-hosted multimodal agentic LLM system on an NVIDIA DGX Spark.
Solar SCADA
  • IoT ingestion for 24 sites at 43,200 data points per second, live.
  • PVLIB and PyBaMM simulations to compare real versus ideal plant performance, QR-based datalogger asset mapping, and health models for each time-series stream.
  • Production deployment on Kubernetes with CI/CD, plus a RAG chatbot for on-demand KPI analysis from live plant data.
Drone inspection
  • U-Net and SAM pipeline that turns high-resolution drone imagery into 3D solar plant layouts, with configurable 3D asset generation and a Three.js viewer.
May 2024 to Nov 2025 · Remote, USA

Deep Learning Engineer, Smarttrak AI

Physics-informed models for solar assets, and my first big VR build.

What I shipped
  • A PINN-based inverter digital twin for predictive maintenance, with a custom-loss PyTorch framework using LSTMs and autoencoders.
  • Short-horizon cloud nowcasting from live NASA Worldview imagery cropped to client sites, with CNN autoencoder cloud classification and feedback-driven retraining.
  • A Unity VR digital twin for Meta Quest with modelled assets, textures, time-of-day lighting, live dashboards and AI chat inside the headset.
2019 to 2024 · Kanpur

IIT Kanpur

B.S. Physics (2019 to 2023) and B.Tech Aerospace Engineering as a second major (2019 to 2024). Computational Physics, Advanced Mathematical Methods, Probability and Statistics, and plenty of LabVIEW and simulation along the way.

Fun bit
  • I got in through JEE Advanced 2019 (rank 4412 out of about 245,000 candidates, in the common rank list).

Selected work, told as problems and results.

Solar SCADA, live

24 sites · 43,200 data points per second

Problem
Operators need to see how every plant is really performing against how it should, in real time.
What I built
IoT ingestion, PVLIB and PyBaMM “ideal plant” baselines, health models per data stream, QR-based datalogger mapping, and a RAG chatbot for KPI questions. Deployed on Kubernetes with CI/CD.
Result
Live monitoring of 24 sites, with actual versus expected performance side by side.

Rooftops from space

30 cm stereo imagery · 1,776 buildings

Problem
Small rooftop objects like HVAC units and tanks decide what fits on a roof, and a flat photo can’t show their height.
What I built
A 3D small-object detection pipeline on stereo satellite pairs, a labelled GIS dataset, and a self-hosted multimodal agentic LLM system on an NVIDIA DGX Spark.
Result
90% functional accuracy on building footprints and heights.

Twins you can walk through

PINNs · Unity · Meta Quest · Three.js

Problem
Asset health is hard to see, and plants are hard to visit.
What I built
A PINN-based inverter twin for predictive maintenance, a Unity VR twin with live dashboards and AI chat, and a U-Net and SAM pipeline that turns drone imagery into 3D plant layouts.
Result
A physics-informed twin plus a walkable VR plant, fed by live data.

Mighty Resume

Solo SaaS · mightyresume.in

Problem
Job seekers rarely know why their resume gets filtered out.
What I built
An AI analyzer scoring nine ATS-focused dimensions, with free and paid tiers and a payment gateway, on Azure with NGINX ingress, TLS, autoscaling and CI/CD.
Result
Designed, launched and operated end to end, alone.

Toolbox

Fresh from GitHub

Pulled live from my public repos when you load the page.

What I think about when nobody’s asking me to ship.

Physics isn’t a line on my CV, it’s how I see things: spacetime bending around mass, information leaving fingerprints in entropy, and intelligence as something we might one day understand as cleanly as gravity.

General RelativityQuantum MechanicsGravitomagnetismInformation entropyKolmogorov complexityAbstract mathAGI timelinesFutures of civilization

The human part.

What it’s like to work with me, minus the buzzwords.

I finish things

Idea, prototype, deployed, monitored. I’m happiest when something I built is actually running out in the world.

I go deep, then wide

One week it’s satellite stereo geometry, the next it’s Kubernetes ingress or a VR headset that won’t track the floor. I get a kick out of crossing the gaps.

I like a good constraint

Tight deadline or tight compute? Fine. Constraints make me pick the elegant solution over the expensive one.

Quietly ambitious

I’ve been better at building than announcing. This site is partly me practising being louder about it.

🏠 Home: Agra🛠️ Base: Gurgaon🔭 Rabbit holes: relativity, entropy, AGI⌨️ Press Ctrl K, there’s a menu

What I’m up to right now.

A “now” page, in the spirit of Derek Sivers. Updated October 2026.

Building

Pushing stereo rooftop detection further, and teaching an agent to write OpenSCAD models of buildings from satellite crops.

Thinking about

How far physics-informed models can go before you need more data, and what AGI timelines really hinge on.

Looking for

A team applying AI to energy, climate or industrial systems, where the physics actually matters.

Let’s make something strange and useful.

Solar, satellites, digital twins, physics, or a good argument about AGI. I’d love to hear from you.

Email me LinkedIn GitHub Hashnode Medium X