Gaurang Khurana // CS @ Waterloo

guest@gaurangkhurana.ca: ~

type help · play

market sim · $1M · 6 crypto · 6 indices · 6 stocks

cash

$1,000,000

portfolio

$1,000,000

P&L

+$0

import portfolio (.csv) drop a file or paste · parsed in your browser, nothing is uploaded

leaderboard · 0 traders · save your run:

no runs yet — allocate, then save your P&L.

crypto = live prices (keyless public APIs, ~20s) · indices & stocks = simulation seeded from real levels, live during each exchange's market hours · +/− = buy/sell · invest your $1M from cash · not investment advice

experience

~/rbc-wm

incoming

Sep 2026 – Dec 2026

RBC Wealth Management

Technical Operation Developer Co-op · Toronto, ON

~/rbc

current

May 2026 – Present

Royal Bank of Canada

Technical Systems Analyst Co-op · Toronto, ON

Built real-time Power BI dashboards integrated with the ServiceNow REST API to consolidate ticket metrics across teams, eliminating ~65% of manual reporting workflows and enabling daily KPI tracking across 10+ service categories.
Provided hands-on technical support across Microsoft 365 (Teams, SharePoint, Outlook, Azure AD) and ServiceNow, resolving incidents and service requests while documenting resolution patterns to reduce repeat ticket volume by ~25%.
Analyzed ticket routing patterns and SLA compliance data to identify resolution bottlenecks, streamlining incident triage and escalation workflows and improving average response time by ~30%.

~/dataknobs

Sep 2025 – Dec 2025

DataKnobs

Software Engineer Intern · Seattle, WA

Built LLM-powered GEO evaluation pipelines (Python, FastAPI) analyzing webpage performance across generative search — automating extraction of entities, claims, and topical relevance and substantially cutting manual review cycles.
Implemented RAG with vector embeddings to ground analyses in verified source content, improving consistency and reducing hallucinations.
Designed backend orchestration + scoring (batch runs, caching, retries, logging) for systematic benchmarking and controlling inference cost at scale.

~/marshalls-consultancy

May 2025 – Aug 2025

Marshall Clinical Research Consulting

Software Engineer Intern · Edmonton, AB

Automated behavioral-health data pipelines (Python, Pandas, FastAPI) for ingestion, cleaning, and structuring — significantly reducing manual prep time.
Spearheaded semantic search via vector embeddings to accelerate literature synthesis and pattern discovery across substances and demographics.
Designed a privacy-first conversational chatbot (LLM APIs, structured prompts, safety rules) for anonymous addiction-awareness and self-reflection.

~/toshniwal-advisory

Jan 2025 – Apr 2025

Toshniwal Advisory Services

Quantitative Trading Intern · Remote

Built a centralized portfolio analytics platform (FastAPI + Next.js) consolidating multi-strategy portfolios — substantially reducing reconciliation time and improving sector-level visibility.
Led sector-exposure decomposition + rotation logic surfacing hidden concentration risk; guiding reallocations that improved returns vs. the prior workflow.
Designed an LLM-assisted analytics layer (Pinecone vector search) to contextualize sector moves, flag abnormal concentration/drift, and accelerate risk discussions.

education

BCS, Computer Science (Honours)

Statistics Minor

University of Waterloo

Elite & President’s Scholarship · $50,000 Group I Distinction · Waterloo Euclid Contest Bloomberg Market Concepts (BMC)

Sep 2024 – Aug 2029

Waterloo, ON

projects

~/pricepilot — main

active

↗ repo

PricePilot latest

Agentic Repricing for Shopify — A Model Proposes, a Deterministic Policy Engine Decides

Go Statistics Shopify Fuzzing Empirical Bayes Elasticity Property Tests

4.5M

fuzz execs
0 violations

94.8%

coverage on a
stated 95% interval

56%

error cut
pooling sparse SKUs

27%

mispricing from one
omitted control

Merchants are right to fear an agent that changes prices, so the guardrails are the product. An elasticity model proposes; a pure deterministic engine approves, clamps, or rejects every proposal against margin floors, MAP, per-cycle caps and repricing velocity — and names the rule responsible every time.
That engine has no clock, network, or randomness, so its six safety invariants are properties a fuzzer can attack rather than claims in a README: 4.5 million executions, zero violations, and the 49 inputs the fuzzer found are committed as permanent regression cases.
The estimator refuses more often than it answers. Elasticity is identified by variation in price, not volume of sales — 300,000 units at one price say nothing about a different one — so thin SKUs borrow strength from their collection and the rest get no proposal at all.
Validated on synthetic demand with a known ground truth, the only setting where recovery can be proven: 95% intervals that cover 94.8% of the time, and one omitted promotion control shown to underprice by 27%. No revenue-lift claim is made — that needs a live A/B on a real store.
Built on BulkPilot’s embedded-app foundation — OAuth, chunked bulk mutations and snapshot rollback reused rather than rebuilt, so week one went to the pricing problem instead of the plumbing.

~/tradebot — main

active

↗ repo

tradebot

Three-Language Trading System — LLM Research, Deterministic Risk Engine, Native Monitor

Go Python SwiftUI Claude API Walk-Forward Monte Carlo MCP

104 ns

risk gate
per order

allocations
race-tested

languages
one pipeline

paper

only — no live
order path

Python researches, Go executes, SwiftUI monitors. An LLM brain proposes trades from indicators and patterns; a deterministic Go risk engine has final say over every order — the model can suggest, but it cannot override position limits, daily drawdown halts, or sizing rules.
Risk gate benchmarked at 104 ns/op with zero allocations and race-tested, alongside the position book and bracket monitor behind an HTTP service.
Backtester with walk-forward optimization and Monte Carlo, plus a TradingView webhook receiver turning Pine alerts into signals. Exposed to Claude over an MCP bridge, so strategies can be swept and charted conversationally.
Ships paper-only by design: no live adapter is wired, and the README is explicit that an edge under ~55% win-rate net of costs never reaches profit — the honest result, not the flattering one.

~/bulkpilot — main

active

↗ repo

BulkPilot

Embedded Shopify App — Bulk Price Editor with One-Click Rollback

GraphQL Admin API App Bridge React Router 7 Polaris Prisma SQLite OAuth

250

variants per
API chunk

1-click

full
rollback

snapshot

every edit
persisted

live

before/after
preview

Bulk-edit product variant prices across a merchant's catalog with a live before/after preview, then undo any change — treating bulk edits like database migrations: preview → apply → roll back. Filter with Shopify's native search (vendor:, tag:), pick an adjustment (set / ±% / ±amount / round to .99), and apply.
Applies changes via the GraphQL Admin API (productVariantsBulkUpdate), grouped by product and chunked to the 250-variant API limit, with per-variant partial-failure handling.
Snapshot-based rollback: each operation persists prior prices via Prisma — rollback replays the snapshots to restore exact pre-edit prices, so a 500-variant markdown gone wrong is reversible in one click.
Built on Shopify's official embedded-app stack (React Router template, Polaris, App Bridge) running against a real Shopify development store.

~/quantcore — main

active

↗ repo

QuantCore

Real-Time Options Pricing & Risk Engine

▸ Live Demo C++ Apple Metal GPU pybind11 FastAPI Next.js WebSocket Monte Carlo

69×

GPU vs NumPy
@ 10M paths

<0.01%

pricing error
vs market

4.4ms

p99 WebSocket
latency

4.5%

VaR breach rate
5% nominal

C++ Black-Scholes + Monte Carlo pricing engine with analytic Greeks, validated to <0.01% against real market option prices via per-strike implied-vol inversion across 335 liquid SPY/AAPL options.
GPU-accelerated Monte Carlo (Apple Metal, Philox PRNG): 69× over vectorized NumPy at 10M paths, 23.5× at the typical 1M-path workload — full host↔GPU transfer included, M3 10-core GPU. CPU path also SIMD + multithreaded (3.8–4.1× over NumPy).
Portfolio VaR backtested on 4 years of real multi-asset data: breach rate 4.5% vs. 5% nominal over 851 days, captured the April 2025 tariff shock.
FastAPI WebSocket streaming to a Next.js scenario-analysis dashboard — p99 4.4ms end-to-end (localhost), validated end-to-end with Playwright.

~/codeshift — main

active

↗ repo

CodeShift

LLM Python 2→3 Migration Tool & Benchmark

parso LangChain Gemini 2.5 Flash Claude Sonnet 4.6 Python

95.1%

pass rate
41 cases

byte≡

Gemini = Claude
pass/fail results

pre-registered
test cases

2/41

shared failures
same root cause

Migration agent: parso-based analyzer (Python 2 can't be parsed by Python's native ast) + LangChain repair loop with a sandboxed test-feedback cycle.
41-case behavioral benchmark with a pre-registered failure taxonomy written before results were collected; every oracle bite-checked to fail on unmigrated source before measurement.
Gemini 2.5 Flash and Claude Sonnet 4.6 produced byte-identical pass/fail on all 41 cases (95.1%), both failures isolating a shared, cross-model inability to emit raise … from None during exception migration.

skills

Languages

Python

TypeScript · JavaScript

Java

C · C++ · C#

Swift

Racket · Lisp · Scheme

SQL · GraphQL · Cypher

HTML / CSS

AI / ML / LLM

TensorFlow · Keras · Scikit-learn

NumPy · Pandas · OpenCV

LangChain · LangGraph

LlamaIndex · RAG

Vector Embeddings · Matplotlib

Finance / Quant

Black-Scholes · Monte Carlo

Greeks · Implied Vol · VaR

Portfolio Analytics · Backtesting

yfinance · Questrade API

Bloomberg (BMC)

Web

React · Next.js · React Router

Node.js · FastAPI · Flask

Shopify Polaris · App Bridge

GraphQL Admin API · Prisma

WebSocket · REST · Django

Electron · SwiftUI

Tailwind CSS · Vite

Systems

C++17 · Apple Metal GPU

SIMD / NEON · Multithreading

Low-latency · Concurrency

pybind11 · CMake

Data & Cloud

PostgreSQL · SQLite

Neo4j · Pinecone · ChromaDB

Redis · Supabase

AWS (EC2) · Azure

Vercel · Render · Fly.io

Cloudflare Workers

Docker · BullMQ

APIs / Integrations

OpenAI · Anthropic · Gemini

Perplexity Sonar · Grok

Twilio · ElevenLabs

Stripe · OAuth · Webhooks

Spotify · VAPI

Tools & Libraries

Git · Bash · Linux

Docker · Cursor · Figma

Postman · JIRA

GitHub Actions · CI/CD

Jest · Playwright · Selenium

D3 · framer-motion · gsap

BeautifulSoup · Requests

github

contribution activity @gkhurana21 ↗

less more