Investment Principal in New York, working across systematic trading and mobility operations. A decade in quantitative finance covering the full lifecycle: alpha research → strategy development → risk management → execution.

• 🌊 Now — Investment Principal — Atlantic Partners: fund distribution & capital introduction, connecting US investors with institutional-quality, EU-regulated strategies across fixed income, alternatives, and equities. Also working on a systematic algorithmic trading system and mobility operations.
• ⚡ Pico (2023 – 2026, NYC) — Engineer on low-latency trading infrastructure and market data systems, working with the Redline Trading Solutions stack: ultra-low-latency feed handlers, order execution gateways, and latency-sensitive production environments supporting institutional clients on execution technology.
• 📊 ACCI Capital Investments (2016 – 2023) — Senior Quantitative Researcher; co-managed multi-asset UCITS funds for pension funds, insurance companies, and family offices. Built deep-learning models for US equity regime forecasting and dynamic allocation frameworks that reduced drawdowns through the 2020 COVID crisis. Led manager selection with full operational and investment due diligence.
• 📈 GAR Investment Managers (2016) — Quantitative Researcher.

MS Data Science — Fordham University · MS Data Science for Finance — CUNEF · MS Applied Mathematics — Universitat Politècnica de Catalunya · Telecommunications Engineering — Universitat Ramon Llull

• reality-explained (private) — an interactive walkthrough of modern physics, information, and consciousness; every claim tagged by epistemic status, null results included · live site
• spotify-popularity-prediction (private) — what makes a song popular? Predicting Spotify popularity from a track's audio characteristics alone, across 56k+ songs (2008–2019), with a 2026 revisit of the original study. The recipe that emerges: danceable, loud, short, darker-toned, with vocals — though sound shapes popularity only at the margins; fame, playlisting, and marketing do the rest · live report
• golf-shot-analysis (private) — 423 of my own 7-iron shots from 9 TrackMan range sessions, run through a Python pipeline that classifies every strike (fat, thin, toe, heel) and ball flight (draw, fade, push, pull) from launch-monitor physics, then benchmarks them against tour-level ranges on an interactive dashboard. The data is humbling and honest — median carry ~60 yds — which is exactly the point: measuring the baseline is how it improves · live dashboard
• wine-tasting-analysis (private) — what does a wine's chemistry actually tell us about its quality? A multi-dataset study of 42,000+ wines that separates the signal that holds up from the claims that don't: alcohol and low volatile acidity track quality, wines fall into a handful of natural style clusters, and chemistry explains only part of what tasters reward. Evaluated on a held-out set under a leak-free pipeline — genuinely useful, but far from deterministic · live report
• dynamic-portfolio-rl (private) — can Markowitz be beaten? A weekly long-short max-Sharpe portfolio with dynamically forecast parameters (ARIMA, SES, VARMA, XGBoost, LightGBM, LSTM, plus an EWMA-LSTM covariance model) head-to-head against a DDPG reinforcement-learning agent allocating across five ETFs through the COVID crash. Static Markowitz won the original 2019–20 test window — but a 2026 rebuild of the research (TF1 notebooks → typed TF2 package) under a strict leak-free, walk-forward evaluation showed the verdict flips when the window ends in the drawdown instead of the recovery: dynamic covariance estimation protects capital exactly when risk shows up. The RL agent compounded +91.7% in its 2018 test period
• facial-tension-detection (private) — computer-vision biofeedback from a webcam: flags facial tension in real time and tracks time spent tense vs. relaxed, so you can notice and unlearn the habit. Privacy-first — no face images are ever stored
• self-peptide space (private) — computational immuno-oncology research in collaboration with a lab at the Icahn School of Medicine at Mount Sinai: can a tumor's mutated peptides be told apart from their healthy wild-type counterparts, and does that difference predict whether the immune system detects them? A rigorous, leak-free analysis lands a clear answer. Among confirmed binders, how tightly a mutated peptide binds the immune system's presentation machinery does not predict whether it gets detected — and neither do the field's favored "altered-self" shortcuts (how foreign or how changed the mutation looks), each tested head-to-head and landing near chance. On this 12-patient cohort, immune detection is only weakly predictable from sequence, sitting just below the frontier of the best published models — independently reproducing the field's hard-won consensus that recognition takes both presentation and context, not any single score. But the story doesn't end on a null: re-running the same honest pipeline on a 9,000-peptide public benchmark lifts the signal from chance to the field's ceiling (~0.63, and ~0.7 with non-linear models), showing this is a data-scale problem, not a cleverer-score one — a smarter feature on 12 patients wouldn't have helped; more data does. Negative results reported rather than buried, and the positive half earned with the same discipline

Alpha Research · Market Microstructure · Low-Latency / HFT Infrastructure · Feed Handlers · Order Execution Gateways · Tick-to-Trade Latency · Co-location · Portfolio & Risk Management · UCITS & Multi-Asset Funds · Manager Selection · Machine Learning · Deep Learning · Reinforcement Learning · Computer Vision · Operations Research & Optimization · AI-Assisted Engineering · Claude Code & Agentic Workflows · Multi-Agent Orchestration

Quant & Machine Learning

Low-Latency & Systems

Product & Web

AI & Agents

📫 Reach me: jtmejon@gmail.com