Skip to main content
JacobJacob Allan — home
Jacob Allan

I build ML andquantitativefinance systems

Machine LearningQuantitative FinanceBackend Systems

Engineering Science student at the University of Toronto — Machine Intelligence and Mathematics, Schulich Leader Scholar. I write Python for machine learning, portfolio risk, and market research, and build the systems around it.

View Selected WorkAbout Jacob

Credentials and current work

02About

Secure systems, careful evaluation, and explicit failure boundaries.

I am an Engineering Science student at the University of Toronto specializing in Machine Intelligence and Mathematics. Most of my work is Python: PyTorch models, portfolio-risk and regime research, evaluation harnesses, and the FastAPI and PostgreSQL services around them — plus deterministic Rust where latency and reproducibility matter.

At Northstar I work in Python across the lifecycle of OdooRedo — application development on Django and PostgreSQL, Documents and IPR workflows, role-based access control and audit trails, rollback-safe data migration, automated verification, CI, and AWS infrastructure provisioned with Terraform.

Education, skills, and approach
01

Python for applied ML

  • Python
  • PyTorch
  • scikit-learn
02

Backend and cloud platforms

  • Python
  • FastAPI
  • PostgreSQL
03

Quantitative research and risk

  • VaR
  • Stress testing
  • Efficient-frontier optimization
04

Systems programming

  • Rust
  • Event-driven design
  • Deterministic replay
7Project and research records
10Work experiences
3Honors and awards
May 2029Expected graduation

03Selected Work

Public systems, with source and limits attached

Public systems and research projects with source, validation, limitations, and working demos where available.

LLM Reliability + EvalOps Platform
LLM EvalOps dashboard showing pass rate, cost, latency, and recent evaluation runs using demo data.

Synthetic/demo evaluation data shown in the public reference implementation.

01 / LLM Evaluation PlatformLive reference implementation

LLM Reliability + EvalOps Platform

A Next.js, FastAPI, and PostgreSQL reference implementation for measuring the quality, reliability, estimated cost, and latency of LLM-powered workflows across versioned datasets, prompts, models, and graders.

Evidence
  • A controlled 20-case RAG regression moved the pass rate from 95.0% to 85.0% and more than doubled estimated cost.
  • 221 backend tests, a successful frontend production build, and passing backend, frontend, and eval-gate workflows.
Stack
  • Next.js
  • FastAPI
  • PostgreSQL
  • Python
  • TypeScript
  • Alembic
  • Docker
  • GitHub Actions
  • Vercel
  • Google Cloud Run
Read case study

More work

Market infrastructure, portfolio risk, local-first tooling, research

Each record carries its architecture, the decisions behind it, the measurements that back it, and what it explicitly does not do.

  1. 02Completed public demoML · Software

    Incident Triage Copilot

    A Next.js and FastAPI application that accepts alerts, logs, metrics, deployment notes, service context, and incident descriptions, then produces a structured summary, severity, impacted service, root-cause hypothesis, immediate actions, and confidence.

    • 16 passing backend tests, Ruff validation, and a successful frontend production build at the evidence snapshot.
    • Public demo verified live and reachable anonymously at incidentcopilottriage.com.
    • Python
    • FastAPI
    • Next.js
    • TypeScript
    • Cloud Run
    Read the Incident Triage case study
    Incident Triage Copilot

  2. 03Open-source systems projectSoftware · Quant · Systems

    Low-Latency Event-Driven Trading Engine

    A Rust event-driven simulation stack centered on price-time-priority market microstructure, deterministic replay, explicit pre-trade risk boundaries, and measured local performance.

    • 247 passing CI tests, deterministic golden-file replay, and property-based tests at the evidence snapshot.
    • Benchmarks are committed as JSON with the machine, OS, and toolchain that produced them.
    • Rust
    • Event-Driven
    • Matching Engine
    • Replay
    • Criterion
    Read the Event-Driven Trading Engine case study
    Concept visual

  3. 04Open-source research platformQuant · ML · Research

    Market Regime + Portfolio Risk Platform

    Builds validated ETF research datasets, estimates portfolio and tail risk, compares multiple regime-detection approaches, runs defined stress scenarios, and evaluates shifted, cost-aware strategies without presenting the output as investment advice.

    • 322 automated checks pass across a Python 3.12–3.14 CI matrix.
    • Chronological splits, train-only scaling, shifted signals, and future-mutation tests guard against look-ahead bias.
    • Python
    • Streamlit
    • Risk Analytics
    • Regime Detection
    • Backtesting
    Read the Market Regime & Risk case study
    Market Regime + Portfolio Risk Platform

  4. 05v1.0.0 public releaseSoftware

    FormatClip

    Uses a minimal-permission Manifest V3 extension, local settings, bounded backend contracts, and optional OpenAI or Groq providers to transform only text the user chooses to send.

    • Tagged v1.0.0 release with a 77,949-byte archive and a published SHA-256.
    • 31 automated tests across the backend and the extension at the evidence snapshot.
    • Chrome MV3
    • WXT
    • React
    • TypeScript
    • FastAPI
    Read the FormatClip case study
    FormatClip

  5. 06Research projectResearch · ML

    RF Signal Classification Research

    Research role

    Developed and compared more than five convolutional architectures, improved noisy-condition accuracy by about 10%, and contributed to a roughly 2 Hz embedded inference pipeline.

    • Approximately 90% average accuracy on unseen real-world signals across five classes.
    • Approximately 10% improvement in noisy-band generalization through augmentation and normalization.
    • PyTorch
    • CNNs
    • RTL-SDR
    • Signal Processing
    • Python
    Read the RF Signal Classification case study
    Concept visual
All projects

04Experience

Software engineering, applied ML research, and quantitative work

Enterprise backend and infrastructure work, RF-signal research, privacy-preserving ML prototyping, and student-led portfolio research.

  1. Jun 2026 – Present

    Northstar

    Software Engineering Intern

    • Python 3.13
    • Django 5.2
    • PostgreSQL 17
    • pytest

    Building secure enterprise backend and applied-ML workflows across OdooRedo, data migration, AWS infrastructure, CI, and end-to-end verification.

    • Built and extended backend workflows using Python 3.13, Django 5.2, and PostgreSQL 17.
    • Worked on controlled Documents and IPR workflows and related RCA/CAPA, approval, notification, external-intake/submission, RBAC, and audit-trail behavior.
    • Implemented and validated rollback-safe migration and reconciliation paths with explicit data-integrity checks.
    • 1,286 PostgreSQL-backed tests
    • 13 Playwright end-to-end tests
    • 9-job CI pipeline

  2. Feb 2026 – Present

    UTEFA

    Portfolio Manager

    Previously Sales & Trading Analyst (Sep 2025 – Apr 2026)

    • Python
    • Pandas
    • NumPy
    • Factor modeling

    Leading portfolio research and risk-aware modeling across factor analysis, portfolio construction, stress testing, and transaction-cost-aware evaluation.

    • Lead portfolio research and review across quantitative and fundamental workstreams.
    • Promoted into this role from Sales & Trading Analyst.
    • Apply probability, factor analysis, Value at Risk, stress testing, and efficient-frontier concepts to portfolio construction.
    • Portfolio research leadership
    • Promoted from Sales & Trading Analyst
    • VaR, stress testing, efficient frontier

  3. Sep 2025 – Aug 2026

    UTMIST / Flybits

    Machine Learning Engineer

    • Python
    • Pandas
    • NumPy
    • scikit-learn

    Developed privacy-preserving customer archetypes from more than 100,000 synthetic personas for a personalized digital-credit-offer prototype.

    • Processed and analyzed more than 100,000 synthetic customer personas.
    • Explored clustering and autoencoder approaches for representation and segmentation.
    • Helped identify more than five interpretable archetypes.
    • 100,000+ synthetic personas
    • 5+ discovered archetypes
    • 6-person team

  4. May 2025 – Sep 2025

    Royal Military College

    Machine Learning Researcher

    • Python
    • PyTorch
    • CNNs
    • NumPy

    Developed and compared PyTorch CNNs for five-class RF-signal classification on approximately 150,000 samples, including noisy-condition and embedded tests.

    • Built and compared more than five convolutional neural-network architectures.
    • Worked with a roughly 150,000-sample, five-class RF dataset.
    • Used RTL-SDR-oriented signal-processing and ML workflows.
    • 5+ CNN architectures
    • 150,000 samples / 5 classes
    • ≈90% unseen-data accuracy
All ten roles, including earlier experience

Core stack

What I build with

The tools that appear across the projects and roles above. The complete grouped inventory, including quantitative and signals work, is on the About page.

  • Python
  • PyTorch
  • scikit-learn
  • XGBoost
  • Pandas
  • NumPy
  • FastAPI
  • Django
  • PostgreSQL
  • Rust
  • TypeScript
  • AWS
  • Terraform
  • Docker
Full technical profile

05Resume + Contact

Software, machine learning, and quantitative systems

One-page resume covering current software engineering, applied ML research, quantitative work, and public systems projects.

Download Master ResumeGet in touch