Skip to content

Halifax, Nova Scotia

Smit Kanpariya

AI EngineerCo-Founder & CTO, Blyyss

Still building. Still learning.

I build AI systems that real people use, and I run my own engineering with AI agents: scheduled, budgeted and gated, with the evals to prove they work. Right now that means running the engineering behind Blyyss, a live grocery platform for independent Halifax shops, and AI research at Saint Mary's University.

Smit presenting Blyyss on stage at Volta in Halifax, with the Blyyss home page on the screen behind him
Demoing Blyyss to 100+ people at Volta, Halifax

01/About

I like knowing how things actually work.

I'm a Computing Science student at Saint Mary's University in Halifax, graduating in December 2027, and I want to spend my career putting AI in front of real users: building it, deploying it, and measuring whether it works.

At Blyyss I co-founded a company and became its CTO at 20. I shipped payments, native mobile apps and a production database to paying customers, and I built the company's engineering around AI agents: scheduled jobs that import 30+ shops' flyers, with change detection, budget caps and a pricing check that blocks a bad publish. Before that I built the retrieval system behind an AI education platform; now I do AI research across three labs.

I enjoy the unglamorous parts: evals before demos, tracing a bug to its cause, making a page fast on an old phone, and writing things down so the next person understands them.

My biggest strength isn't knowing everything.It's knowing how to keep learning.
Curious, careful, and comfortable saying “I don't know yet”.

02/Experience

Where I've been learning.

Each role taught me something I couldn't have learned from a course. Here is what I worked on, and what stayed with me.

  1. May 2026 – Present

    Halifax, NS

    Co-Founder & CTO (from Sep 2026); Founding AI Engineer (May – Aug 2026)Blyyss Technologies Inc.

    I co-founded Blyyss with our CEO and lead everything technical: product engineering, web and mobile, backend, payments, data and testing.

    • Built and run blyyss.ca and its iOS and Android apps: 140+ orders delivered to 90+ customers, rated 4.7/5.
    • Turn requirements from the CEO, store owners and our delivery partner into features: delivery fees, order cut-offs, payments and privacy controls.
    • Run the engineering with AI agents: scheduled headless LLM agents import 30+ shops' flyers behind change detection, budget caps and a publish gate; 800+ merged pull requests, backed by 50+ unit test suites and 380+ Playwright end-to-end tests.
    • LLM agents
    • React
    • TypeScript
    • Supabase
    • PostgreSQL
    • Stripe
    • Capacitor
    • Playwright

    What it taught meShipping to real customers teaches faster than any tutorial, and it punishes guesses.

  2. Sep 2026 – Present

    Halifax, NS

    Research Assistant, Artificial IntelligenceSaint Mary's University

    Research across three artificial intelligence projects with three Saint Mary's University professors.

    • Dr. Jiju Poovvancheri's lab (SPG-JEPA): scaling Point-JEPA, the lab's self-supervised model that learns 3D shapes from unlabelled point clouds, from single objects up to city-scale scenes.
    • With Dr. Ethan Pancer: applied AI research at the intersection of large language models, agentic AI and machine learning, built on Blyyss's live retail data.
    • With Dr. Somayeh Kafaie: using TWOSIDES side-effect data to compare a simple drug network with a multigraph that keeps each side effect, to see whether the extra detail changes which drugs look most important.
    • Self-supervised learning
    • 3D point clouds
    • LLMs
    • Agentic AI
    • Network science

    What it taught meA good question matters more than a clever model.

  3. Jan 2026 – Apr 2026

    Halifax, NS

    Data Analyst Intern (Co-op)Saint Mary's University — EIT

    Modernized how the university's reports are built and delivered.

    • Moved legacy reports to Power BI Report Builder and deployed them to Microsoft Fabric and Azure.
    • Wrote Oracle SQL to structure reporting data and automate recurring reports.
    • Power BI
    • Microsoft Fabric
    • Azure
    • Oracle SQL
    • Excel

    What it taught meA report is only useful when the people reading it trust the numbers.

  4. May 2025 – Dec 2025

    Remote

    Founding AI EngineerKAbot Robotics

    Early engineer on kabot.ca, an AI platform that helps teachers plan, assess and track their classes.

    • Built the RAG system behind the platform's teacher dashboards from scratch, with Gemini as the model, Qdrant as the vector database and Docker for deployment.
    • It grounds lesson planning in the curriculum and powers AI quizzes, learning-gap detection and student insights, alongside the gradebook, attendance and calendar.
    • RAG
    • Gemini
    • Qdrant
    • Docker
    • FastAPI
    • Python

    What it taught meRetrieval decides the answer. Most of the work is choosing what the model gets to read.

  5. Sep 2024 – Aug 2025

    Halifax, NS

    Math TutorVarsity Tutors & SMU Students' Association

    Helped students understand mathematics and work through problems.

    What it taught meIf I can't explain something simply, I don't understand it yet.

03/Projects

Things I've built, and what they taught me.

Each one started with something I didn't know how to do yet.

Featured · Live productMay 2026 – Present · Co-Founder & CTO since Sep 2026

Blyyss

Order from your local Halifax shop at the price on its shelf. A grocery storefront, flyer app and delivery service for independent stores, on the web, iOS and Android.

orders delivered
140+orders delivered
customers
90+customers
average customer rating
4.7/5average customer rating
local shops' flyers
30+local shops' flyers

Featured on Global News · CTV News · Radio-Canada · The Coast · Country 103.5

A full room at Volta in Halifax watching the Blyyss demo

Demo night at Volta, Halifax: walking a room of 100+ people through Blyyss, live.

Innovation Sprint at the Arthur L. Irving Entrepreneurship Centre, September 19, 2026: Blyyss was the partner company, and 20 students from four universities spent 90 minutes tackling real Blyyss challenges, then pitched their solutions.

Blyyss home page showing this week's flyer prices from Gateway Meat Market
Blyyss flyers page listing local shops' weekly flyers, nearest first

The problem

Halifax has dozens of independent grocers and almost none of them can be ordered from online. Customers who want those stores have to get there, and most of the shops have no website to tell anyone what is on their shelves.

What I built

  • A storefront where a customer orders from one local store's real catalogue at the store's own shelf price, and we shop and deliver it.
  • Tappable digital flyers from 30+ independent Nova Scotia shops, with search across every flyer and nearest-store sorting.
  • An admin dashboard for analytics, order tracking, payment settlement and planning each delivery day.
  • One React and TypeScript codebase that ships as the website, an iOS app and an Android app.
  • An agent-native engineering process: scheduled headless LLM agents import flyers and update the catalogue, behind change detection, per-run budget caps and a publish gate that checks every price against the live pricing function.

Technical decisions

The server is the authority on money

After a security review, every order is re-priced on the server from its own copy of the catalogue. A tampered or out-of-date price is refused instead of charged.

Show the most you can pay, before you pay

Stripe places a hold for the maximum possible total at checkout and only captures the real amount after the groceries are bought, so a weighed item can never surprise anyone.

Phone first, measured

Most sessions are on phones, so layouts are tested automatically at real phone widths, and the product catalogue is kept out of the first page load.

Agents do the repetitive work, gates decide what ships

Flyer imports and catalogue updates run as scheduled LLM agents with a budget cap per run and a lock file. Nothing an agent produces reaches customers unless it matches the pricing function the server uses to charge a card, so a wrong model output costs a retry, never a wrong price.

Tests as a safety net

50+ unit test suites, 380+ Playwright end-to-end tests and type-check gates for the web app and all 21 server functions are what make shipping often to real customers possible.

Case studyAugust 2026 · Debugging a live product

The orders that vanished after our Facebook launch

sessions replayed
~150sessions replayed
left at sign-in
70–80%left at sign-in
the fix
6 digitsthe fix
  1. The symptom

    Payments, frontend and backend all worked. Then we started marketing on Facebook, and orders dropped sharply even though people kept visiting and engaging.

  2. The investigation

    I set up Microsoft Clarity session recording and reviewed about 150 real sessions, following click patterns and time spent. 70–80% of visitors were abandoning at the sign-in step.

  3. The root cause

    Facebook opens links in its own in-app browser. Our sign-in emailed a magic link, and tapping it opened the phone's default browser instead: a second, separate browser, so the customer was never signed in where their cart was.

  4. The fix

    I replaced the link with a six-digit one-time code typed straight into the page, so sign-in completes inside Facebook's browser, and hid Google sign-in there because Google blocks it in embedded browsers. Orders came back.

Nothing was broken, and we were still losing most of our customers at the door. As a co-founder I learned that every detail, big or small, is the business.

What I learned

Real customers turn every shortcut into a support message. I learned to treat money, data and edge cases as product features, and that the fastest way to learn a technology I didn't know (payments, native mobile, serverless) is to ship it carefully to people who depend on it.

  • React
  • TypeScript
  • Supabase
  • PostgreSQL
  • Stripe
  • Capacitor
  • Playwright
  • GitHub Actions
  • Vercel

The code is private. I'm happy to walk through it in a conversation.

Bailey's Meat Market website prototype: Nice to Meat you
Local Source Market website prototype: In store this week

Websites for local grocers

Seven website prototypes for independent shops in Halifax, Dartmouth and Nova Scotia

Problem
Most independent grocers around Halifax have no website, or one that can't tell a customer what is in the store this week.
What I built
A site for each shop in its own identity: weekly deals and flyers, what's in store this week, hours and directions, and English and Bengali side by side for a Bangladeshi grocer.
Technical decision
Each site is designed around how that shop's customers actually shop (a halal meat counter, a farm market's seasonal tables, a fish guide) instead of one template with the logo swapped.
What I learned
Design starts with the customer. A butcher and a farm market need different websites, even when both sell groceries.
  • Next.js
  • React
  • HTML
  • CSS
  • Responsive design
  • Vercel
Woodland Conservation Area website hero

Woodland Conservation Area

A public website for a conservation foundation

Problem
The Woodland Conservation Foundation needed a site that shows off the woodland and that future student developers can keep improving.
What I built
On a team of five, I built the Supabase sign-in and likes data layer and page features of a responsive site with interactive maps and text-to-speech audio.
Technical decision
The team added Cypress end-to-end tests and a step-by-step handover guide, so the next students inherit a working foundation instead of a mystery.
What I learned
Writing for the next developer is part of the job: tests and documentation are features too.
  • React
  • TypeScript
  • Vite
  • Leaflet
  • Supabase
  • Cypress

Always learning

I build things I don't fully know how to build yet.

Technology changes constantly. I try to stay curious, pick projects slightly past what I know, and treat every one as a chance to understand something better.

  1. 2023Started Computing Science at Saint Mary's University in Halifax.
  2. 2024Began tutoring math, and found that explaining something is the fastest way to find what I don't understand.
  3. 2025Built the RAG system behind KAbot's teacher dashboards as a founding AI engineer.
  4. 2026Data analyst co-op, then joined Blyyss as its founding AI engineer and became Co-Founder & CTO in September, learning payments, native mobile and testing by shipping them.
  5. NowAI research across three labs, an agent-native engineering process at Blyyss, and the next step: shipping AI that customers use directly, with the evals to prove it works. Still building.

04/Skills

Tools I've worked with.

These are technologies I've used in real projects, not ones I claim to have mastered. Most of them I'm still getting better at.

AI engineering
  • LLM agents
  • RAG
  • Prompt engineering
  • Gemini
  • Claude
  • Qdrant
  • Evals and publish gates
  • Claude Code
  • Codex
Languages
  • Python
  • TypeScript
  • JavaScript
  • SQL
  • HTML
  • CSS
Frontend
  • React
  • Next.js
  • Tailwind CSS
  • Vite
  • Leaflet
  • Recharts
Backend & APIs
  • Supabase Edge Functions (Deno)
  • FastAPI
  • REST APIs
  • Stripe API
Databases
  • PostgreSQL
  • MySQL
  • Oracle SQL
Mobile
  • Capacitor
  • iOS
  • Android
Cloud, Data & DevOps
  • Vercel
  • Docker
  • GitHub Actions
  • Microsoft Fabric
  • Azure
  • Power BI
Testing & Tools
  • Playwright
  • Cypress
  • Git
  • GitHub

05/Contact

Let's build something, or talk about what you're building.

I'm open to co-op and internship roles, and always happy to hear from people working on interesting problems.