Montréal, QC
Forward Deployed Engineer (Co-op)
May 2026 – Present
•Delivered a self-serve video-ad studio for a leveraged-ETF issuer as primary backend author (26 of 28 commits) by shipping a FastAPI + Next.js portal with Creatomate templates, ElevenLabs voice cloning, and Kling image-to-video on S3/ECS.
•Screened every user-edited ad script against SEC/FINRA Rule 2210 before render by pairing a deterministic rules pass (banned claims, required disclosures) with a second LLM reviewer pass, logging every verdict to a Postgres audit trail with CSV export.
•Cut a 30-second render from 6 generated clips to 2 and enabled 15–30s shots from a single still by routing each request to the Kling model with the longest single-call length and chaining last-frame image-to-video with ffmpeg.
•Closed an SSRF path on user-supplied media URLs by adding a URL guard that blocks private and metadata addresses before any server-side fetch.
•Turned a SharePoint archive of support calls into searchable, speaker-labelled transcripts for a Japanese industrial pump OEM by building an idempotent AssemblyAI → S3 pipeline and an ERP unit-history integration (invoices, RMAs, part swaps) feeding the team's hybrid RAG knowledge base.
•Gave field technicians an in-app reference of 29 product articles (~175k words) without bloating the client by splitting ~1 MB of prose out of the Next.js bundle at build time.
•Automated support-ticket routing to 5 owning teams by building a TypeScript webhook service that runs Desk365 tickets through an AI tagger (0.65 confidence floor) and opens Asana tasks with custom fields, plus a backfill script and 8 tests.
•Resolved high-priority client AI incidents within a 24-hour SLA by triaging with stakeholders and shipping patches, rolling Medium/Low fixes into bi-weekly releases.
•Moved a follow-on engagement from discovery to scoped work by writing PRD inputs and leading weekly PMOs and client demos, while mentoring an intern through tasking and code review.
Full-Stack Engineer
May 2025 – May 2026
•Automated $250M+ in investment transactions across 2,000+ subscription agreements for a $1T+ AUM asset manager by building Emma (Outlook intake, fund classification) and Oslo (field extraction, human-in-the-loop review, versioned audit trail).
•Reduced per-document handling 93% (30 min to under 2 min) at 99.37% extraction and 98.98% classification accuracy by owning full-stack delivery on both agents and running evals on scanned documents.
•Took the vision LLM out of checkbox and signature reads (0 wrong answers on a 13-document eval) by measuring ink density in OCR-anchored Yes/No regions, backed by 126 new tests.
•Corrected page orientation on 36/36 rotated test cases (vs 13/20 for Tesseract OSD) and cut extraction time ~25% (96–121s to 69–93s) by rebuilding rotation detection and reading e-signatures straight from PDF data.
•Raised old-vs-new form detection from 7/10 to 10/10 and 5/11 to 11/11 across two fund types by adding a tiered text-layer → OCR fallback, keeping all 710 extraction tests green.
•Diagnosed from 270 production emails that 21.1% missed the 30s hand-off budget (p95 604s) and 8.9% double-submitted, then fixed both with a Postgres advisory lock and submission triggered on every classification outcome.
•Gave ops and executives live routing-accuracy, trade-volume, and processing-time visibility by building the Oslo KPI dashboard (29 of 31 commits) in React/TypeScript with Recharts, filters, and one-click PDF export.
•Authored ~2/3 of the Oslo review workspace and ~3/4 of the Emma inbox by shipping document versioning, audit logs, discard reasons, investor notifications, and in-browser PDF/Excel review.
Frontend Developer
Nov. 2024 – May 2025
•Created a 10+ component React/TypeScript library reused across 20+ agent workspaces by standardizing UI primitives, Next.js parallel routes for zero-reload switching, and Recharts KPIs; accelerated client sign-off with prototypes and MVPs.
Self-Employed
Montréal, QC
Freelance Software Developer
Sep. 2024 – Present
•Delivered 5+ production web applications with responsive, accessible, and SEO-optimized interfaces, directly supporting client acquisition and user engagement.
•Achieved zero-downtime deployments across all client projects by implementing CI/CD pipelines on Vercel with automated testing.
•Eliminated manual booking and communication overhead by automating client workflows through GoDaddy and Gmail API integrations.
•Surfacing where prediction markets disagree on the same event, measured as percentage-point divergence across 2 exchanges (Kalshi, Polymarket), by normalizing both into one market model, matching equivalent markets, and computing divergence, movers, and consensus in Python.
•Keeping AI chat answers grounded in market data, with 0 probabilities or margins generated by the LLM, by having Python calculate every figure and AWS Bedrock only explain the results.
•Cutting exchange API calls per chat query to 0 by serving every answer from Foretell's own Postgres, fed by REST ingestion into a current-state markets table plus 5-minute historical snapshots (Docker locally, RDS in production).
•Shipped a word game with 4 modes to production, validating guesses against 245k words and serving 596 curated answers, on FastAPI/Next.js/PostgreSQL with pre-deploy Alembic migrations on Railway.
•Built server-authoritative 1v1 Versus over WebSockets (1 Hz HP drain, deadline-checked guesses, no repeated words across rematches), covered by 175 pytest tests.
•Hardened the public API against abuse with per-IP sliding-window rate limits (300 writes/min), request-size caps, a WebSocket Origin allowlist, and CSP/HSTS headers.
•Curated 734 clean definitions from Free Dictionary and Wiktionary by building an ingest pipeline that filters answer-leaking or truncated glosses and grades difficulty by word frequency.
•Reached 1.36 test RMSE, 6.9% better than a grid-searched Random Forest (1.46), predicting upvotes at publish time by fine-tuning DistilBERT with 12 metadata features on 225k posts from a 450k-post dataset.
•Benchmarked 5 models (Dummy, Ridge, Random Forest, title-only and content-aware BERT) by streaming the compressed Reddit dump without loading it into memory, with score-stratified sampling and fp16 training.
•Added real-time multiplayer to a GeoGuessr-style campus game and migrated it from MongoDB to Supabase Postgres with SQLAlchemy/Alembic (102 files changed), authoring 21 of 53 commits.
•Built an open-source Agent Meter for a macOS menu-bar app that tracks live usage, plan limits, and spend for Claude, Codex, and Cursor across accounts by parsing local session logs without storing prompts (+2.4k lines of Swift, with tests).
Certifications
Amazon Web Services
AWS Certified Cloud Practitioner
Amazon Web Services
Languages
AI / ML
Frameworks
Cloud / Data
Bachelor of Computer Science
Expected Dec. 2026
Relevant Coursework: Data Structures & Algorithms, Machine Learning, Data Analytics, Operating Systems
Champlain College, Saint-Lambert, QC
DEC in Commerce
Sept. 2020 – May 2023
Relevant Coursework: Micro & Macroeconomics, Financial Accounting, Business Law, Quantitative Methods