Janav Dhingra

Software developer & data scientist

I used to move the data. Now I ask it questions.

  • Developer
  • Tinkerer
  • Traveller

Hi, I'm Janav. I spent three years as a software engineer. Somewhere along the way I noticed I knew exactly where the data went, but not what it was saying. So now I'm doing a Master of Applied Data Science at the University of Victoria to close that gap.

NowDoing my master's at UVic. Open to data scientist, ML engineer, AI engineer and software developer roles.
now

Hover a year to see what I was doing, or drag the dot.

Janav in the gardens outside the Empress hotel in Victoria

Work

What I've been doing, newest first.

01 University of Victoria Master of Applied Data Science Jan 2026 – now

The switch.

For three years my job was making data move faster. I got one request from 228 milliseconds down to 45 and felt great about it.

Then I noticed the forecasts I was shipping around were more interesting than the shipping. So I went back to school to work on the part that thinks.

Same data. Better questions.

Coursework in statistical learning, machine learning, optimization, data warehousing and data visualization, plus a research project on how far you can trust a model's predictions (it's in Projects). I'm also in the UVic Data Science and Entrepreneur clubs.

02 for Software Engineer Oct 2024 – Dec 2025

Turning demand forecasts into what goes on the shelf.

I led development of an internal stock and demand system for Walmart North America. The Data Science team sent forecasts in, and the system worked out what each store needed and where it should go: shelf, hanger or warehouse. Stores in coastal and dry climates got different advice, depending on how fast items were selling.

60%less time to generate vendor plans
80%lower average response time
3×the throughput

Java · Spring Boot · PostgreSQL · Kafka · Redis · AWS

Infosys Insta Award, 2025 for this system.

03 forApple Software Engineer Aug 2023 – Sep 2024

Making the vendor numbers agree with each other.

My first project after university. Apple's vendor tracking reports didn't always add up, so I designed a multi-step SQL workflow that fixed the inconsistencies, and a pipeline of scheduled batch jobs and event-driven processing to feed it. I also pulled reports and one-off extracts from large transactional datasets for the business side.

SQL · PostgreSQL · Spring Boot · 85%+ test coverage

04 Oneistox Software Development Intern Dec 2021 – May 2022

Where it started.

My first engineering job, during my B.Tech in New Delhi. I reworked MongoDB schemas and queries so data loaded 30% faster, and built the REST APIs and backend for onboarding, course tracking and mentoring.

Node.js · Express · MongoDB

Projects

Things I built because I wanted to.

Featured · Mobile web app

EDIT: Daily Style

One outfit a day, picked from your own closet. Nobody needs an infinite scroll at 7:40 in the morning.

EDIT checks the live weather where you are and what's on your calendar, then gives you one main look, two backups, and a short reason why it works. A presentation gets something sharper. A lecture gets something easier.

The picks come from a scoring system I wrote by hand. It rates each combination on colour harmony, fit balance, formality and warmth, and it learns from what you like, save and actually wear.

  • No accounts, no server database. Everything stays on your phone.
  • Installs like an app. It's a PWA, so it lives on your home screen.
  • Calendar sync through a small serverless relay on Netlify.

React · JavaScript · Open-Meteo API · iCal parsing · Netlify Functions · browser storage

EDIT's Today screen: 14° and clear in Victoria, a 4 PM lunch on the calendar, and the pick: light blue oxford, grey knit sweater, navy shorts
EDIT's outfit card listing why the outfit works, with other options below
EDIT's closet screen showing a grid of 25 shirts and sweaters

Today's pick · Why it works · Your closet

Janav in a mirror selfie wearing the outfit EDIT picked: light blue oxford shirt, grey knit sweater over the shoulders, navy shorts
What it picked, worn.

Master's research · Machine learning

How sure is the model? Prediction intervals for jet engines

A model that says "74 more flights" sounds confident. One that says "58 to 85, guaranteed 90% of the time" is actually useful.

With three classmates at UVic, I worked on predicting how many flights a turbofan engine has left before it's unsafe, using NASA's C-MAPSS data. We wrapped LSTM and gradient-boosting models in seven conformal prediction methods, which turn a single guess into a range with a coverage guarantee, and tested that guarantee across 112 combinations of data, model and method.

It held almost everywhere. Where it broke, on the hardest dataset and only for mid-life engines, we used Kolmogorov–Smirnov tests to trace it to calibration and test engines not looking alike, then repaired it with weighted CQR. Then we broke it on purpose with a frozen-pipeline stress test, just to see how it fails.

Python · LSTM & quantile LSTM · gradient boosting · conformal prediction (MAPIE) · KS tests · 63 unit tests

Point model: "74 more flights" Conformal: "58 to 85 flights, 90% guaranteed" 58 85 40 100 flights
Same model underneath. The range is what an engineer can safely act on.
15.30
LSTM error (RMSE) on FD001, under the 16 target
30%
narrower intervals with CQR, same guarantee
89.5%
coverage restored on the one dataset that broke
1
command to reproduce every table and figure

Full-stack · Microservices

Pixagram

I built this before AI was writing everyone's code. I kinda like it.

An Instagram-style app where you can post photos, like, comment, follow people and see what's trending. Six Spring Boot services sit behind one gateway that checks your login token on every request, and each service keeps its own MySQL database. One docker-compose up starts all seven containers.

Trending hashtags get pulled out of captions as you post. The Caption Studio writes captions in four tones and tags every post with one of seven vibes, all from hand-written keyword lists and templates, because that's what AI meant back then.

Angular 17 · TypeScript · Spring Boot · Spring Cloud Gateway · MySQL 8 · JWT · Docker Compose

See it on GitHub →
6
microservices
4
database schemas
17+
REST APIs
7
containers, one command

Tools

What's in my requirements.txt

Everything you'd need to reproduce me, minus the coffee.

requirements.txt42 dependencies · 0 conflicts
  1. 1# machine learning & ai
  2. 2PyTorch # Neural networks for the master's.
  3. 3TensorFlow # The other half of the deep learning coursework.
  4. 4scikit-learn # Statistical learning and machine learning, from the coursework at UVic.
  5. 5MAPIE # Conformal prediction intervals for the jet engine research.
  6. 6Hugging Face # Transformers, mostly. Pretrained models are a gift.
  7. 7LangChain # Wiring language models up to tools and documents.
  8. 8Claude # My go-to AI pair for thinking things through and writing code.
  9. 9Codex # OpenAI's coding agent, for when I want a second pair of hands.
  10. 10MCP # Model Context Protocol: lets AI assistants plug into real tools and data.
  11. 11
  12. 12# data & analysis
  13. 13Python # Most of the master's happens here.
  14. 14pandas # The first import in every notebook I open.
  15. 15NumPy # The second import in every notebook I open.
  16. 16Jupyter # Where most experiments start, and some of them end.
  17. 17MATLAB # Matrices first, questions later.
  18. 18Tableau # Dashboards for people who don't want to read SQL.
  19. 19Power BI # Same idea, Microsoft flavour.
  20. 20Excel # Still the world's most popular database. I've made peace with it.
  21. 21LaTeX # The jet engine paper was written in it. Looks better, takes longer.
  22. 22
  23. 23# databases, backend & cloud
  24. 24PostgreSQL # Fixed Apple's vendor reporting with a multi-step SQL workflow.
  25. 25MySQL # Behind Pixagram, one database per microservice.
  26. 26SQLite # The database that's already on your phone.
  27. 27MongoDB # Made queries 30% faster at Oneistox.
  28. 28Redis # Took Walmart's average response time from 228 ms to 46 ms.
  29. 29Kafka # Event streams for Walmart's stock and demand system.
  30. 30Node.js # REST APIs at Oneistox, my first engineering job.
  31. 31AWS # ECS, Lambda and S3 on the Walmart project.
  32. 32Docker # Daily. It works on my machine, and now on yours too.
  33. 33
  34. 34# languages, web & engineering
  35. 35Java # Two years of production backend at Infosys.
  36. 36Spring Boot # What the Walmart and Apple systems ran on.
  37. 37JavaScript # EDIT, among other things.
  38. 38React # Components all the way down. EDIT is built on it.
  39. 39Angular # Pixagram's front end. React's more opinionated cousin.
  40. 40C # Learned during my B.Tech in New Delhi.
  41. 41C++ # Also from the B.Tech, along with a healthy respect for memory.
  42. 42LabVIEW # Programming by drawing wires between boxes.
  43. 43Siemens # Industrial automation and PLCs.
  44. 44
  45. 45# tooling & staying organized
  46. 46Git # Daily. Commit messages are getting better.
  47. 47GitHub # Where my side projects live.
  48. 48pytest # Tests that read like plain sentences.
  49. 49YAML # Config files, and the occasional indentation crisis.
  50. 50Jira # Tickets, sprints, and the occasional “won't fix”.
  51. 51Todoist # Where my to-do list lives, and occasionally gets done.
$ pip install janav
Successfully installed janav-2026.10. Hover a package to see why it's here.

Off the clock

When I log off.

For a while I spent a little too much of my life in front of a screen. My code was in great shape. Everything else, less so.

So now I log off on purpose. I travel whenever I can, take photos of most of it, and try to actually live a bit.

I also produce and mix music, which is mostly signal processing with better snacks.

Janav in a red shirt leaning on a railing, looking out across a glass-roofed atrium

Lessons so far

What broke, and what it taught me.

From Apple

If two reports disagree, assume they're both lying.

On the vendor tracking platform the bugs weren't in the code, they were in the numbers. I now trust data the least when it looks the most finished.

From

Fast is easy. Fast and correct is the job.

Multi-threading made vendor plans 60% faster. Stopping two vendors from overwriting each other at peak time was the actual work.

From the master's

AI is a power tool, not a pilot.

I use it to move faster on the parts I already understand, then spend the time it saves building tools for myself. Exhibit A: EDIT, the app that picks my outfit every morning so I don't have to.

From life

AI can plan the trip. You still have to go.

It'll write a decent itinerary in ten seconds. It can't get lost in a new city, find the good street food, or take the photo. That part's still on me, and it's the best part.

Contact

Say hello.

I'm happy to talk about data science and software development roles, a project you're working on, or music. I reply to email quickly. Not 45 milliseconds quickly, but close.

LinkedIn GitHub
Janav in a tan sweater holding a coffee, in a long plant-lined walkway