Ad astra, per aspera
AI & Automation · Portfolio
Aryan
I build ML systems that scale, adapt, and actually make sense. Scroll to follow the route — from the foothills of the Himalayas to the West.
Chapter 01 · Descent
Third planet from the Sun.
Sunlight takes 8 minutes 20 seconds to get here. Every route needs a first point — mine is Dharamshala, in the foothills of the Himalayas.
Chapter 02 · The route
Six stops.
One line connects every place I’ve studied and worked. Keep scrolling and it draws itself.
01 Dharamshala Born & schooled 1999–2017 02 Noida B.Tech, Amity University 2017–21 03 Gurugram TechOps Engineer, Comviva 2021 04 Montreal Master’s, Concordia University 2023–25 05 Longueuil Associate I, Amazon YUL5 2023–25 06 Montreal AI & Automation Manager, CSC Generation 2025–
Waypoint 01 / 06
Dharamshala
Himachal Pradesh, India
Born & schooled
1999 – 2017
Where it all started. Born and raised in the foothills of the Dhauladhar range, and where I went to school.
Origin Home base
Next waypoint Noida · 420 km south-east
Waypoint 02 / 06
Noida
Uttar Pradesh, India
B.Tech, Computer Science & Engineering
Amity University
Jul 2017 – Jun 2021
Leg 01 · 420 km From Dharamshala
Next waypoint Gurugram · 31 km west
Waypoint 03 / 06
Gurugram
Haryana, India
TechOps Engineer, Comviva
Jun – Aug 2021
Java + Spring enterprise apps serving 500+ users
REST microservices: 40% more scalable, 25% faster responses
Leg 02 · 31 km Across Delhi NCR
Next waypoint Montreal · 11,271 km
Waypoint 04 / 06
Montreal
Québec, Canada
Master of Applied Computer Science
Concordia University
Aug 2023 – May 2025
Leg 03 · 11,271 km Over the Arctic
Next waypoint Longueuil · 6 km
Waypoint 05 / 06
Longueuil
Québec, Canada
Associate I, Amazon Canada
YUL5 · Oct 2023 – May 2025
Resolved 100+ device issues daily, keeping 95% uptime
Scripted routine tasks: 15% fewer manual errors
Log monitoring cut incident response time by 30%
Leg 04 · 6 km Across the St. Lawrence
Next waypoint Back to Montreal
Waypoint 06 / 06 · Now
Montreal
Québec, Canada
AI & Automation Manager
CSC Generation / Sur La Table · Oct 2025 – Present
PO automation (n8n, Gemini, Microsoft Graph): 50–100 orders a day, zero manual entry
6+ MCP servers as one-click .mcpb installers, used by ~200 people across CSC brands
Led the company-wide Claude rollout and “Claude Day” (March 2026)
Montreal · where I workToronto · registered office, 235 Bay StreetSeattle · Sur La Table HQ
Chapter 03 · In orbit
Projects, launched.
Five missions orbiting home. Select a satellite to read its telemetry.
All
AI / ML
Web
Mobile
Mission 01 AI / ML
Tik AI
TikTok AI Agent
Next.js 14 React 18 TypeScript Tailwind CSS Gemini 2.5 Flash Veo 2/3 Vertex AI AI-powered content analysis with Google Gemini 2.5 Flash at 90%+ accuracy and under 2s response time, plus Veo 2 video generation for 5–8s promotional clips.
Telemetry
Next.js + TypeScript dashboard supporting 100+ concurrent users
Modular AI services handling 1,000+ video assets
Docker + Vercel deployment at 99%+ uptime
Mission 02 Web
Scrapy
TikTok scraper with AI vector search
Bun Hono BullMQ Redis Playwright transformers.js TypeScript Docker Scraped 100+ TikTok videos with full metadata, then made them searchable with all-MiniLM-L6-v2 embeddings — 100ms to embed, under 50ms to search.
Telemetry
Architecture handling 1,000+ concurrent embeddings
BullMQ + Redis queue: 3 jobs every 5s with 120s timeout protection
About 2KB of storage per video
Mission 03 AI / ML
Multi-Agent Chatbot
LLM chatbot with reinforcement learning
Python Llama 3 LangChain FAISS FastAPI Streamlit RLHF A multi-agent chatbot with 92% routing accuracy and 87% response relevance. FAISS made RAG context retrieval 3× faster.
Telemetry
RLHF lifted engagement 35% and cut latency 45%
Reward model trained on 1,000+ feedback logs for adaptive re-ranking
Streamlit dashboard for live analytics and feedback
Mission 04 Web
TubeLytics v2
Reactive YouTube analytics dashboard
Java Scala Play Framework Akka Actors WebSockets Docker GitHub Actions Streams channel, video and tag analytics from the YouTube Data API v3 over WebSockets — no refresh needed.
Telemetry
Akka supervision + request caching cut API calls by 60%
Non-blocking word-frequency and tag modules over 1M+ words per query
90%+ unit-test coverage with JUnit 5 and sbt CI
Mission 05 Mobile
OpenTracks
Open-source Android sport tracker
Java Android SDK Gradle Material Design GitHub Actions New UI features and voice announcements for an open-source sport tracking app, for better usability and accessibility.
Telemetry
Aggregate-statistics dashboard: distance, duration, pace and elevation across all tracks
“Announce average slope” toggle wired into the existing TTS pipeline
220 LOC across 5 core classes, legacy announcements intact