Hi, I'm Pravdeep. I'm a forward deployed engineer and full-stack developer. Welcome to my website. I work with React, Next.js, TypeScript, Node.js, Python, Claude API, RAG, AI agents, FastAPI, Firebase, PyTorch, scikit-learn, and whatever the problem needs.
Forward Deployed Engineer at Nx Growth Partners. Previously a full-stack developer at Mentera, an AI platform for private practices and medspas. I hold a graduate certificate in AI with machine learning from Humber.
I also move small businesses off slow, high-maintenance WordPress sites onto fast custom sites that don't need plugin updates.
Pravdeep Kullar
Forward deployed and full-stack engineer
Written by me, not a model.
The problems I like are the ones where the model isn't the hard part.
I'm a Forward Deployed Engineer and full-stack developer with over 2 years of hands-on experience in React, Next.js, TypeScript and Node.js, plus Python on the ML side. At Nx Growth Partners I work directly with client teams to scope AI and software problems into clear milestones and ship them. Before that I was a full-stack developer at Mentera, an AI platform for private practices and medspas, where I built features for the product, including the Canvas panel for Tera, its assistant, which renders the artifacts Tera streams back. Earlier I interned at Rebelsnow building client e-commerce sites.
I hold a graduate certificate in AI with machine learning from Humber, which was mostly useful for knowing what a model can and can't be asked to do. I have working Claude API integrations for text and vision, and I use Claude Code and Cursor every day.
Outside of work I build things I want to use myself. A multiplayer Wordle clone with real-time rooms, auth, and a socket server I deployed myself. A tool that reads a photo of my class timetable and hands back calendar events. A Raspberry Pi that watches Cineplex for ticket on-sale times. Countdown widgets for the iOS Home Screen. Making the interface not feel slow is usually the interesting part.
Twelve projects. Some were paid, the rest I built because I wanted the thing to exist.
Nx Growth Partners is a custom software and AI development agency. I work directly with client teams to understand the business problem, scope it into a clear first milestone, and coordinate NxGP's senior engineers through build and delivery. The work is AI agents, copilots, RAG and knowledge systems, custom web and mobile apps, and integrations into the stack a client already runs.
Mentera is an AI platform for private practices and medspas. Tera is the assistant inside it, and staff lean on it in the middle of a working day. I built the Canvas side panel, where the artifacts Tera streams back (markdown, code, HTML, SVG, Mermaid and React) render as they arrive instead of as a wall of chat text. HTML and React run in a sandboxed iframe, and the assistant needed backend changes to support it, which I made alongside the backend owner. I also built the config-driven integration detail drawer that covers 35 integrations.
Content arrives in pieces, and drawing a piece before it's finished makes the panel flicker. Each block now waits until it's whole before any of it renders. The code that does the waiting is ugly. The flickering stopped.
Two of the sites I've worked on had the same problem: a heavy hero video in front of everything else. On the Next.js marketing site for Mentera I compressed the hero video, migrated images to next/image, deferred GTM, switched to next/font/local, and lazy-loaded 21 Framer Motion components. On Rosehill the video was large enough that I compressed it with ffmpeg and added a Cloudflare R2 fallback for delivery.
The Mentera pass targeted 75+ on desktop and 65+ on mobile PageSpeed.
Rosehill is a design-build contractor in Ontario. Their old site was built in 2013 and looked unfinished, and on a phone it never loaded at all. I fixed that, and together we built a new Next.js marketing site on Vercel, with me handling the domain and DNS setup. We edited a new hero video together. It got ffmpeg compression and a Cloudflare R2 fallback, and it loads after the rest of the page so nothing waits on it. I added OG tags so sharing a link stopped looking broken, plus GTM and cookie consent, and ran the feedback rounds with the client's COO. I'm still working on it today, because I believe in getting it perfect.
The owner had no interest in learning a CMS, so changes came to me through a Notion board and I made them myself. That was faster for both of us, and it means a new phone number still has to come through me.
Rising Sun is a WASC-accredited Montessori school in El Dorado Hills, California, teaching transitional kindergarten through middle school. The people reading the site are parents choosing a school, so the navigation is built on the four programs instead of on the school: each grade band has its own page, and the campus and gallery pages are there because parents want to see a classroom before they agree to visit one.
Enrolment doesn't happen on the site. Applications hand off to Transparent Classroom, which the school already runs, and a tour is booked by emailing the office, because tours are Wednesday mornings only and one person keeps that calendar. Two of the three things a parent came to do finish somewhere I don't control. Building either one on the site would have meant keeping a second copy of records the school already has.
I interned at Rebelsnow from January to July 2024, building and shipping client e-commerce websites from the client's requirements through to deployment. That meant the pages customers land on and the parts they buy through: the catalog, the cart, checkout and payment.
Checkout got the most testing, because it's the one place where a bug costs the store a sale instead of just looking wrong.
fakewordle.com is Wordle, except you play it against someone else. You open a room, send someone the code, and race them in real time with a chat running alongside. Accounts and past games are saved in Firebase Auth and Firestore, Claude Haiku generates the words, and it connects to Google and Outlook calendars over OAuth.
I ran the live game straight off the database to begin with, which meant a write for every letter a player typed, and you could feel the delay. Moving the match onto its own live server fixed it, and the database only stores the result now.
Every semester I typed my class timetable into a calendar by hand. Now I upload a photo of the timetable, Claude vision parses it, and I get back Google Calendar and Outlook events.
The model was the easy part. A timetable is a grid where position carries the meaning: where a class sits tells you when it is, so parsing the text alone throws away half the information.
A bot that monitors Cineplex for movie ticket on-sale times and sends a push notification so I can grab tickets quickly. It runs always-on and headless on a Raspberry Pi 4 with Raspberry Pi OS Lite (64-bit), managed over SSH.
An iOS app with countdown widgets for the Home Screen. The app is Expo SDK 54, React Native 0.81, React 19, TypeScript and Expo Router. The widgets are SwiftUI with WidgetKit and AppIntents, and a local native Swift module handles the storage the app and widgets share. I build and sideload it through Xcode.
One of my two capstones. It takes health data and predicts risk, and the point of the project was putting a plain statistical model up against a small neural network on the same data to see what the extra complexity actually bought.
I prepared the data and ran the comparison between the two models, then built the interface for it, because a page of numbers in a notebook is not something you can put in front of anyone outside the program.
Fake job postings are a real problem for people applying for work in Canada. HuntigoX checks the employer behind a listing before anyone applies: it confirms the business is real and registered, then scores the posting against the patterns that show up in scams. The compliance side is built around Canadian hiring rules.
I built what candidates and employers actually see, including the verification steps and the scorecard at the end of them. On the model side I worked on spotting the patterns that mark a listing as suspect.
Humber Polytechnic, January to August 2026. Model work run locally on an M4 MacBook through the MPS backend.
Fully connected variational autoencoder on CIFAR-10. The torchvision mirror was crawling, so I wrote a parallel downloader and pulled the archive myself. The reconstructions came back as coloured blur, which is what a fully connected encoder does to 32x32 colour images. The assignment was about the VAE and not the architecture, so I left it and wrote up why a convolutional encoder would fix it.
ResNet18 cat and dog classifier with best-epoch checkpointing and MPS support, plus a set of image processing labs.
Toxic comment detection on the Jigsaw dataset for the midterm, using SVM over Word2Vec and Doc2Vec embeddings. Earlier work covered named entity recognition, TF-IDF, and LSTM sentiment analysis.
A VAR and Hive assignment, a GAN in TensorFlow and Keras generating synthetic patient vitals, and a Thompson Sampling lab.
Market basket analysis with Apriori, and seasonal decomposition on time series. Group presentation reviewing the ReCANet paper on repeat consumption in recommender systems.
Feb 2026–now
Forward Deployed Engineer
2026
Freelance Web Developer
Jan–Aug 2026
Graduate certificate, artificial intelligence with machine learning
Jul 2024–Dec 2025
Full-Stack Developer
Jan–Jul 2024
Web Developer Intern
Sep 2023–Aug 2025
Diploma, computer programming
Got something you want built?
I'm open to full-time front-end and full-stack roles, and I move small businesses off WordPress onto fast custom sites. If the idea isn't figured out yet that's fine, I'd rather hear it early. Email is the fastest way to reach me.