Hi, I'm Yash Sah

CS @ Waterloo. National ICPC champion. Incoming Co-op Researcher @ Huawei Noah's Ark Lab (Fall 2026).

More than just code.

I'm a CS student at Waterloo (Faculty of Mathematics Global Scholarship, President's Scholarship of Distinction). Right now I split my time between coursework and building Datastream, a market-data platform at Wat Street, and this fall I'll be doing ML/CV research at Huawei's Noah's Ark Lab.

Before Waterloo I grew up in Doha, Qatar, where I won the ICPC Qatar national championship, founded my school's first inter-school hackathon, and picked up a Shotokan black belt and a Trinity College keyboard certification along the way.

UWaterloo CS 2A
ICPC National Champion
ML / AI Builder
Doha → Waterloo
7 Projects
4 Hackathons
16 Awards

Everything, start to finish.

Sept 2026 – Apr 2027
Incoming · Research Co-op

Huawei Noah's Ark Lab

Incoming Co-op Researcher · 8-month term · Markham, ON

Joining Noah's Ark Lab for an 8-month co-op as a Research Co-op focused on machine learning and computer vision. More here once I start.

Machine Learning Computer Vision Research
May 2026 – Present
Design Team · ML Engineer

Wat Street (Datastream)

ML Engineer · Datastream Team

I'm one of three engineers building Datastream, a distributed time-series platform for Wat Street, Waterloo's quantitative finance design team (~40 members). Before Datastream, every project on the team pulled market data through its own ad-hoc yfinance scripts. We're replacing that with a single ingestion and serving layer that 5+ downstream projects query instead. I work across the stack: FastAPI services, Postgres storage, the Svelte frontend, and Docker deployment.

FastAPI PostgreSQL Svelte Docker Python Time-Series
Apr – May 2026
Internship · Cloud Infrastructure

Mannai Corporation

Cloud Infrastructure Intern · Doha, Qatar

I deployed a private cloud from nothing: two bare-metal Oracle servers to a working VMware alternative (Platform9 PCD on k3s + OpenStack) with a self-service dashboard where internal teams provision their own VMs, networks, and storage.

"From nothing" meaning the first server didn't even boot cleanly. I traced a misleading I/O error to a missing MegaRAID logical volume, rebuilt the array from 6 drives, and stood up Ubuntu, KVM, and OpenStack on top. Later, when every HTTPS request from every tenant VM started dying, I chased a TLS handshake failure down to a blanket DNAT rule misrouting port 443 return traffic. I also built a custom Ubuntu + Kubernetes image with Cluster API tooling for on-demand cluster provisioning, and onboarded a second server (768 GB RAM) that refused every normal install path until I served the ISO over NFS as a virtual CD-ROM from the first machine.

OpenStack Kubernetes k3s Cluster API KVM/libvirt Bare Metal OVN Linux
2026
Case Competition · Semifinalist

UW Management Consulting Case Competition

Semifinalist · Top 8 of 80 teams

Reached the semifinals with a $1B ESG capital-allocation framework for Saudi Aramco, modelling ~1.5 MtCO₂e/year verified abatement within 36 months at a ~16% portfolio IRR. Conducted primary research via Bloomberg Terminals and IEA net-zero pathways; built a MAC curve with 10-year NPV sensitivity benchmarked against EU ETS. Defended to judges from McKinsey, BCG, Deloitte, and KPMG.

ESG Financial Modelling Bloomberg NPV MAC Curve Strategy
2026
Case Competition · Round 2

Google Build with AI Case Competition

NeuroAdapt · Advanced to Round 2

Built NeuroAdapt · an AI exam prep platform that reframes learning as a diagnostic problem. Classifies mistakes as conceptual, procedural, computational, or fatigue-driven to isolate exact prerequisite gaps. Each mechanism is grounded in peer-reviewed cognitive science: Bjork's New Theory of Disuse, Kapur's productive failure (d = 0.6–2.3), and Gollwitzer & Sheeran's implementation-intentions meta-analysis. Defended to Google engineers and case judges.

Next.js Tailwind CSS Supabase Framer Motion AI Cognitive Science
2026
Hackathon · 1st Place

DDC × Ignition Hacks 2026

Urban Marble · First Place Winner

Led a 4-person team to build Urban Marble · an AI-powered 3D city exploration platform. Features a mobile-first AR building scanner (live webcam frames + GPS → GPT-4o Vision → architectural metadata), a cinematic story pipeline (GPT-4o-mini narration + DALL-E 3 imagery + Google Street View + OpenAI TTS, all parallelised under 10s), and a 400+ rental listings aggregator across NYC and Toronto. Built on a CesiumJS 3D globe with OSM Buildings geometry.

Next.js TypeScript Cesium.js GPT-4o-mini DALL-E 3 Framer Motion
Feb 2026
Hackathon · Winner

Hack Canada 2026

Build with AI Track Winner · Finalist: Most Technically Complex Hack

Built PulmoScan · an ML-powered lung disease detection and 3D visualisation platform from raw DICOM CT scans. Engineered a dual-track pipeline combining Otsu thresholding with a fine-tuned EfficientNet-B0 classifier and Grad-CAM heatmaps, paired with an interactive VTK.js 3D lung viewer using marching cubes surface extraction. Named finalist for Most Technically Complex Hack and currently advancing to Google's startup funding track.

Python PyTorch React Next.js VTK.js FastAPI EfficientNet GradCAM
Dec 2025 – Jan 2026
Internship · Data & AI

Qatar Financial Centre Regulatory Authority (QFCRA)

Data & AI Intern · Banking Supervision Team

Built a Graph-RAG intelligence system (FastAPI, PostgreSQL + pgvector, Gemini 2.5-flash) for semantic retrieval across 80+ regulatory and geopolitical events, enabling natural-language supervisory queries with inline source citations. Developed a news monitoring pipeline tracking 60+ financial entities and their related parties across 8 data sources with automated risk classification across 8 taxonomies, cutting signal identification from hours to minutes. Enhanced Tableau dashboards consolidating risk indicators across 60+ QFCRA-authorised entities, reducing monthly/quarterly prudential return review time by ~70%.

FastAPI PostgreSQL pgvector Graph-RAG Gemini 2.5 Tableau KNIME Python
Sept 2025 – Present
Education

University of Waterloo

Bachelor of Computer Science · GPA 3.9/4.0 · Class of 2030

Studying CS with a focus on algorithms, machine learning, and systems programming. Holder of the Faculty of Mathematics Global Scholarship ($15k CAD) and the President's Scholarship of Distinction ($2k CAD).

Algorithms C Python Linear Algebra Systems
2024
Competition · 1st Place

ICPC National Champion

Qatar Collegiate Programming Contest · 300+ participants · 6,000 QAR prize

Won 1st place among 300+ participants in Qatar, scoring equivalent to the 2nd-place university-category team. Advanced to the regional Arab Collegiate Programming Contest (ACPC). Featured in the Gulf Times.

Competitive Programming Algorithms C++ ICPC ACPC
2024
Leadership

Core Organiser · DPS Hackathon

First Inter-School Hackathon in Qatar · 100+ participants

Designed and led the school's inaugural inter-school hackathon with 100+ participants, fostering collaboration and innovation in STEM fields. Brought in Cisco professionals as judges. Featured in the Gulf Times.

Event Management Leadership Qatar Gulf Times
2024
Competition · Finalist

Alice3 Game Dev Competition Finalist

CMU Qatar · Top 80+ teams

Reached the final level of the Alice Programming Competition held at Carnegie Mellon University Qatar, competing against 80+ teams. Developed skills in 3D animation and event-driven game programming.

Alice3 3D Programming CMU Qatar Event-Driven
2024
Olympiad

Zonal Computing Olympiad (ZCO)

Indian Computing Olympiad · Top 100

Obtained a top-100 finish in the first stage of the Indian Computing Olympiad, qualifying through the Zonal Computing Olympiad round.

Competitive Programming Algorithms C++
2023–24
Certification

Cisco Python Certifications

Python Essentials 1 & 2 · Skills for All

Completed Cisco's official Python Programming certifications (Python Essentials 1 and 2) through the Skills for All programme.

Python Cisco Certification
2020–2024
Olympiad · Multiple Medals

High School Olympiad Medals

Silverzone & SOF · Grades 9–12

Earned multiple medals across STEM olympiads organised by Silverzone and SOF (Science Olympiad Foundation) throughout high school, competing across subjects including Mathematics, Science, and English at school, regional, and international levels.

Silverzone SOF Mathematics Science

You've reached the end of the beginning.

Congratulations, you just scrolled backwards through five years. There is nothing below this but bedrock.

Things I've built.

QFCRA Supervisory Intelligence Suite
  • Graph-RAG · Built a Graph-RAG intelligence system (FastAPI, PostgreSQL + pgvector, Gemini 2.5-flash) that ingested 80+ regulatory and geopolitical events, enabling natural-language supervisory queries with inline source citations across QFCRA's monitored institution universe.
  • News Pipeline · Developed an AI-powered news monitoring prototype tracking 60+ financial entities and their related parties across 8 data sources, with an automated classification layer categorising signals into 8 risk taxonomies aligned with QFCRA's supervisory framework, cutting identification time from hours to minutes and reducing manual triaging by ~5 hrs/week per analyst.
  • Tableau · KNIME · Processed 5 regulatory and market datasets in KNIME (management bulletins, QFCRA licence registry) into clean input tables, then enhanced Tableau dashboards consolidating risk indicators across 60+ QFCRA-authorised entities, enabling supervisors to complete monthly/quarterly prudential return reviews in 2 hours (70% reduction).
  • Research · Collaborated on a cybersecurity and AI adoption risk assessment across 60+ QFC-authorised entities, synthesising survey responses and assessment visit findings into a 10-page briefing with institution-level risk heat mapping. Produced 2 analytical summaries linking global geopolitical events to QFC entity exposure using an AI-assisted pipeline that cut production time by 75%.
Python FastAPI PostgreSQL pgvector Graph-RAG Gemini 2.5 Tableau KNIME
NeuroAdapt

AI exam prep platform that reframes learning as a diagnostic problem. Classifies mistakes as conceptual, procedural, computational, or fatigue-driven to isolate exact prerequisite gaps. Grounded in Bjork's New Theory of Disuse, Kapur's productive failure (d = 0.6–2.3), and Gollwitzer & Sheeran's implementation-intentions meta-analysis. Defended to Google engineers and case judges.

Next.js Tailwind CSS Supabase Framer Motion TypeScript OpenAI SDK
Saudi Aramco ESG Framework

$1B ESG capital-allocation framework for Saudi Aramco modelling ~1.5 MtCO₂e/year verified abatement within 36 months at a ~16% portfolio IRR and ~$88/tCO₂e marginal abatement cost. Bloomberg Terminal research, MAC curve with 10-year NPV sensitivity benchmarked against EU ETS. CFO-led stage-gate governance. Defended to judges from McKinsey, BCG, Deloitte, and KPMG.

ESG Financial Modelling Bloomberg NPV MAC Curve Carbon Pricing
Connect Four AI

Connect Four engine with minimax and alpha-beta pruning. Plays optimally at depth 8 in under 50ms.

C Python Algorithms
Health Data Visualizer

Interactive dashboard that ingests patient CSV data and renders real-time visualisations for vitals, medication history, and trends. Built in Python with MySQL and Matplotlib.

Python MySQL Matplotlib
Smart Dots

Genetic algorithm simulation where dots evolve to reach a goal, visualised in real time. Tweakable mutation rates, population sizes, and obstacle layouts.

JavaScript Canvas API Genetic Algorithms
Datastream

Distributed time-series platform serving market data to Wat Street, Waterloo's quantitative finance design team. Replaces the ad-hoc yfinance scripts each project used to maintain with one ingestion and serving layer that 5+ downstream projects query. FastAPI services over Postgres, a Svelte frontend for exploring series, and Docker deployment. Built by three engineers.

FastAPI PostgreSQL Svelte Docker Python Time-Series
Private Cloud on Bare Metal

A self-service cloud for internal teams at Mannai: provision your own VMs, networks, and storage from a web dashboard, on hardware that started as two bare Oracle servers. Platform9 PCD on k3s over OpenStack, as a VMware alternative. Getting there meant rebuilding a MegaRAID array from 6 drives, tracing a site-wide TLS failure to a DNAT rule misrouting port 443 return traffic, and baking a custom Ubuntu + Kubernetes image with Cluster API tooling for on-demand cluster provisioning.

OpenStack Kubernetes k3s Cluster API KVM/libvirt Bare Metal Linux

What I work with.

Python Python TypeScript TypeScript C C C++ C++ JavaScript JavaScript React React Next.js Next.js PyTorch PyTorch FastAPI FastAPI Git Git Docker Docker Node.js Node.js Tailwind CSS Tailwind CSS Three.js Three.js Supabase Supabase PostgreSQL PostgreSQL Linux Linux NumPy NumPy SciPy SciPy MySQL SQL / MySQL R R Cesium.js Cesium.js VTK.js Framer Motion Framer Motion OpenAI SDK OpenAI SDK Matplotlib Matplotlib Tableau Tableau RAG / Vector Search REST APIs Financial Modelling Geospatial / 3D Computer Vision Transfer Learning Grad-CAM DICOM / Medical Imaging SimpleITK WebGL Multi-Agent AI Vercel Vercel Anthropic SDK Anthropic SDK Gemini SDK Gemini SDK Groq SDK LangChain LangChain Google Maps API Google Maps API OpenStreetMap OpenStreetMap Zustand shadcn/ui shadcn/ui Radix UI Radix UI Recharts Pygame Cobe Lucide Lucide KaTeX KaTeX Mermaid Mermaid jsPDF react-markdown react-markdown

Want to build something?

I'm always open to interesting projects and conversations. Let's talk.