ShubhOS
A PORTFOLIO YOU CAN OPERATE
drag the windows · try the Terminal · everything is clickable
Welcome

Welcome to ShubhOS.

I'm Shubh Pathak — I build AI systems and data platforms at UBS Investment Banking. Instead of telling you that on a long page, I built you a small operating system. Everything here works:

Experience — my work at UBS, before & after
Neural Lab — train a real neural network, live
Match — paste your JD, get an honest fit report
Terminal — for the engineers (try sudo hire shubh)
About Shubh
SP

Shubh Pathak

Open to new opportunities · Pune, India · happy to relocate

I make enterprise data do the work — AI systems that read 8,000 documents a month, models that decide 4,000 approvals a year, and dashboards that watch 80+ systems. Built over 2.5 years at one of the world's largest banks, from intern to owning a generative AI platform end-to-end.

System specifications
OSShubhOS 3.0 (human, reliable)
ProcessorPython · SQL · DAX
GraphicsPower BI · Microsoft Fabric
AI acceleratorAzure OpenAI · LangGraph · TensorFlow
Memory24 certifications (MS · Google · Oracle)
Uptime2.5 yrs @ UBS, zero crashes
EducationB.Tech CS, VIT · 8.70 CGPA
Experience — 2.5 yrs @ UBS
Projects — personal builds
Impact Monitor

Numbers, not adjectives.

Live readout of what my systems handle in production.

0+documents auto-reviewed every month
0+ hrsmanual work eliminated monthly
0%accuracy of my approval model
0+decisions automated per year
0+enterprise systems, one view
0certifications · MS, Google, Oracle
14 days → 3 days−67% API calls−70% latency2 FTEs saved
Neural Lab — train one yourself

Teach a neural network. Right now.

Machine learning is my field: systems that learn from examples. Here's one, written from scratch in this page's source. Drop two kinds of dots, press train, watch it learn.

0epochs
loss
accuracy

1 · Give it examples

Click the canvas to place dots — or load a pattern:

2 · Let it learn

The shaded regions are the network's belief. Watch it bend around your dots.

Under the hood

Two hidden layers (12 + 12 neurons), hand-written backpropagation, ~60 lines of dependency-free JavaScript. Like TensorFlow Playground, it also sees simple transformations of your coordinates (sin, x·y, r²) — that's what lets it crack the spiral. The same idea, scaled a few billion times, powers what I build at work.

Match — JD screener

Paste your job description.
I'll do the screening.

My day job is building systems that read documents and check them against requirements — so here's one that reads yours. Runs entirely in your browser; nothing is uploaded or stored.

or try a sample: ·
0%

Résumé — 2 files

Two lenses, one engineer.

Same career, framed for different readers. Take whichever fits your role — or both.

PDF
Shubh_Resume_AI_Engineer.pdf
LLM platform, contextual routing, async pipelines, Ada
PDF
Shubh_Resume_Data_Analyst.pdf
INSPIRE dashboard, ETL pipelines, BI delivery, stakeholder work

Not sure which? Open Match, paste your JD, and it'll recommend one.

Terminal — shubh@portfolio
shubh@portfolio:~$
Contact

Your data has work to do.
Let's talk.

Exploring roles where enterprise data meets automation — India or anywhere in the world.

GitHubshubh0614
LocationPune, India · open to relocation
Email me now