Tahsin Fatin — Portfolio©2026

ঢাকা থেকে মন্ট্রিয়ল

Duke of
Dhaka

Loading

Soundtrack — 23 Theme (From “AA23”) · Anirudh Ravichander

0%

Hello! I'm

Tahsin
Fatin

A Data-Driven

BuilderAnalyst
Selected works ↓
GitHub ↗

01 — About Me

Hello, again

IgrewupinDhaka—acityoftwentymillionpeople,infinitetraffic,andbetterfoodthanwhereveryou'rereadingthisfrom.AteighteenIlandedinHalifaxforafinancedegreeatSaintMary's,whereImanagedareal$600KstudentfundandtooksecondplaceattheVentureCapitalInvestmentCompetitioninBoston.

Thencamesixyearswherefinanceanddatakeptcolliding:fundoperationsatCitco,thenRBCCapitalMarkets,whereIgrewfromDataAnalysttoSeniorDataAnalyst—buildingprojectionmodels,automatingreportingpipelines,andcuttinga13-minutedatapipelinedownto2.SomewhereinthereIalsobecameaCFALevel1candidate.

NowI'minMontréaldoingaMasterofManagementinAnalyticsatMcGill'sDesautels,goingdeeponthemachine-learningside:NLPonfinancialnews,computervisionforCanadianNationalRailway,andwhateverelseletsmebuildthingsthatmatter—someofthemforBangladesh.

Currently
MMA candidate @ McGill (Desautels)
Previously
Senior Data Analyst @ RBC Capital Markets
Based in
Montréal, Canada
From
Dhaka, Bangladesh
Credential
CFA Level 1 candidate

02 — What I do

What
I Do

Analyze

Description

Six years of making capital-markets data confess: projection models, SQL at scale, NLP on financial news, computer vision on live rail footage. If it has a signal, I'll find it.

Build

Description

Then I ship the answer: automated pipelines that cut 13 minutes to 2, production ML behind FastAPI and Docker, and DeshRide — a whole carpooling platform for Bangladesh.

03 — My CareerMy career & experience

  1. Master of Management in Analytics

    McGill University — Desautels Faculty of Management, Montréal

    2025 — present

    Awarded a 30% entrance scholarship. Projects: forecasting stock moves from news sentiment (ARIMA + BERT), predicting speed-dating success (GBM, k-NN), and an ongoing partnership with Canadian National Railway building real-time computer vision that flags rail-line deformities and wildfire risk from live video.

  2. Senior Data Analyst

    RBC Capital Markets, Halifax

    2024 — 2025

    Built projection models and validation reporting on large relational databases; automated data-mining workflows that halved report generation time; optimized a core data pipeline from 13 minutes down to 2.

  3. Data Analyst

    RBC Capital Markets, Halifax

    2022 — 2023

    Shipped a big-data Python program that went to production, and reported weekly to senior management through Tableau dashboards.

  4. Operations Analyst

    Citco Fund Services, Halifax

    2021

    Fund services for multinational hedge funds and private equity — A/B, sensitivity and stability testing, client-facing decks, and a 20% ETL efficiency improvement from a data-validation investigation.

  5. Research Associate → Fund Manager

    Impact Fund, Sobey School of Business

    2018 — 2020

    Research Associate (2018–19), then Fund Manager (2019–20) of the TMT book in a $600K student-run fund, valuing companies with DCF/DDM/comps on FactSet, Bloomberg and Capital IQ. Pitched the Maxar liquidation and the OpenText acquisition — both executed. Plus co-ops at Nova Scotia Power (energy forecasting) and East Coast Offshore Supplies.

  6. Bachelor of Commerce, Finance

    Saint Mary's University, Halifax

    2016 — 2020

    GPA 3.88/4.30, Magna Cum Laude, Dean's List 2017 & 2018, Beta Gamma Sigma (top 7%). Placed 2nd at the Venture Capital Investment Competition in Boston. Graduated with Co-op Distinction.

04 — Works

My Work

Every number below is quoted from the project's own README.

The brand, from the repo
The brand, from the repo
01Flagship

64

districts covered

DeshRide

সমগ্র বাংলাদেশ — intercity carpooling for Bangladesh

A Poparide-style carpooling platform for all of Bangladesh. Drivers post trips they're already making; travellers book the empty seats. Payments sit in escrow via bKash, Nagad, or card until the trip completes, designed around Bangladesh Bank's digital-commerce rules. Ships as an Android app with automated APK builds.

TypeScriptViteCapacitorEscrow paymentsGitHub Actions

View the repo ↗

Fig. Coordination outcomes by scenario
Fig. Coordination outcomes by scenario
02MGSC 697 · capstone

13/13

adversarial scenarios pass

Portfolio Committee

A governed multi-agent investment committee

Macro, Sector, Risk and Compliance agents bid through a risk-budget auction, with veto controls and a human CIO approver. Governance fails closed: unsafe or ambiguous decisions escalate instead of trading. Tested against prompt injection, stale data, restricted assets, replay, over-budget bids and specialist timeouts.

Multi-agent systemsOpenAI Agents SDKPythonEvals

View the repo ↗

Fig. Growth of $1 — Elastic Net vs SPY, 2017–2023
Fig. Growth of $1 — Elastic Net vs SPY, 2017–2023
03FINA 695

0.948

Sharpe — vs 0.633 for SPY

Walk-Forward Alpha

ML models for cross-sectional stock-return prediction

Linear, tree-based and neural models compared walk-forward over 84 out-of-sample months (2017–2023), each driving a monthly long-short portfolio. Elastic Net delivered a 14.83% annualized long-short return and 17.41% alpha versus SPY.

Elastic NetGradient boostingTensorFlowWalk-forward CV

View the repo ↗

MAXR → SELLOTEX → BUYDLB · EXPO · MKTX
Calls & ideas
04Impact Fund · Saint Mary's

2/2

IC pitches executed by the fund

Equity Research

From a student fund's IC to moat-hunting

At the $600K Impact Fund I built the Maxar SELL case — DCF (perpetuity and exit multiple) checked against comps — and pitched the OpenText buy; the fund executed both. Lately: three small/mid-cap moats — Dolby's licensing standard, Exponent's reputation, MarketAxess's network effect — valuation and risks included.

DCFComparable companiesMoat analysisBloomberg · FactSet

No public repo

Fig. Grad-CAM — where the network looks
Fig. Grad-CAM — where the network looks
05MGSC 695

93.5%

CIFAR-10 test accuracy

Residual CNN + Grad-CAM

A custom vision model, and proof of what it learned

A 7.6M-parameter residual CNN whose activation and receptive field were picked through a controlled four-way ablation, transferred from CIFAR-10 to CIFAR-100 (72.35%), then opened up with feature maps and Grad-CAM.

PyTorchComputer visionTransfer learningGrad-CAM

View the repo ↗

Fig. Normalized confusion matrix
Fig. Normalized confusion matrix
06MGSC 695

77.6%

macro-F1 — vs 71.6% TF-IDF

Transformer Topic Classifier

Fine-tuned RoBERTa vs a classic baseline

Multi-class topic classification on 20 Newsgroups: RoBERTa fine-tuned and calibrated with temperature scaling, benchmarked against TF-IDF + logistic regression, then exported to ONNX with perfect prediction agreement.

RoBERTaTransformersONNXCalibration

View the repo ↗

live video→ real-time inference→ defect & wildfire flags
Pipeline
07

CN

industry partner · in progress

Rail Vision × CN

Real-time computer vision for Canadian National Railway

Ongoing McGill partnership with CN (NYSE: CNI): live video analytics that detect rail-line deformities and flag conditions that start track-side wildfires — models built for real-time inference on streaming footage.

Computer visionPythonReal-time inference

No public repo

Fig. Mean profit per 28-day episode
Fig. Mean profit per 28-day episode
08

95.2%

service level — vs 92.0% heuristic

Café Inventory RL

Should a DQN run the kitchen? An honest audit

Q-learning, DQN and PPO against a business heuristic for daily replenishment, stress-tested through demand spikes, collapses, supplier disruption and spoilage shocks. The verdict was governance, not hype: keep the heuristic live and shadow the DQN for four weeks.

DQNPPOGymnasiumDecision science

View the repo ↗

request → agent plan→ guardrail re-audits→ approved or blocked
Control flow
09

8/8

offline evals pass

Course Advisor Agent

An AI agent that can't break the rules

A course-planning agent on the OpenAI Agents SDK whose guardrail never trusts the model's own risk claims: it recomputes the policy audit and blocks altered risks, wrong approval flags and blocked-course recommendations.

OpenAI Agents SDKGuardrailsPydanticpytest

View the repo ↗

notebook → package→ FastAPI → Docker→ Render + drift watch
Pipeline
10INSY 674

API

FastAPI · Docker · Render

ML in Production

From notebook to deployed API

End-to-end ML predicting whether job candidates will change employers: research notebooks → production package → FastAPI with Pydantic validation → Docker → live on Render with drift monitoring. My contribution hardened the API contract with CI tests.

scikit-learnFastAPIDockerCI/CD

View the repo ↗

05 — Skillset & Tools

My Techstack

Analytics & Machine Learning

  • Python
  • SQL
  • pandas / scikit-learn
  • Deep learning (PyTorch · TensorFlow)
  • NLP (RoBERTa · BERT)
  • Computer vision
  • Reinforcement learning (DQN · PPO)
  • Time series (ARIMA)
  • Tableau

Finance & Markets

  • Valuation (DCF · DDM · Comps)
  • CFA Level 1 candidate
  • FactSet · Bloomberg
  • Capital IQ · Reuters
  • Fund operations
  • Quantitative research

Engineering & Tools

  • Agentic AI (OpenAI Agents SDK)
  • FastAPI · Pydantic
  • Docker · ONNX
  • pytest & evals
  • TypeScript / React
  • Git & GitHub Actions
  • ETL pipelines · SAS

06 — Beyond the résumé

Life, lately

ঢাকা → Halifax → Montréal

Dhaka gave me the hustle, Halifax gave me a finance degree and my first real winters, Montréal gave me bagels and machine learning. Three cities, one wardrobe crisis.

The soundtrack

The song playing right now is the 23 Theme by Anirudh Ravichander. It's what I put on when it's 2 a.m. and the model finally converges. Non-negotiable site feature.

Hear it on YouTube ↗

Red Cross roots

Before the spreadsheets: coordinating for the Red Cross. Some instincts — show up, organize the chaos, help — carried straight into how I work.