Open to full-time · Jan 2027 · NYC

9w[xV* qZSu;Data Scientist

I build AI systems for finance, analytics and
real-world decisions.

Data scientist, AI engineer and quant. I work on multi-agent LLM systems, forecasting and trading research — taking ideas from a research notebook to something people use in production.

See the work

Try live apps Ask my AI

status.logonline

Now: Data Scientist – Quantitative Research (Part Time) @ Man Group × Columbia
Studying: MS Data Science · Columbia University
Based in: New York, NY
Focus: LLM systems · forecasting · quant research
Latest: AI at the Margin

Graduating December 2026

70%less time to extract contract termsJain Global · Quant Developer Intern

12–18%reduction in client churnPaperchase · AI Engineer

$10Binstitutional portfolio simulatedBMLL Technologies × Columbia

40%fewer manual forecast adjustmentsPaperchase · AI Engineer

50KBloomberg data fields analysed for costJain Global · Quant Developer Intern

20%better forecast accuracy (MAPE)Paperchase · AI Engineer

30%wider anomaly-detection coveragePaperchase · AI Engineer

1stAI adoption strategy case competitionColumbia University · 2026

#85in the US, IMC Prosperity 4Trading competition · 2026

70%less time to extract contract termsJain Global · Quant Developer Intern

12–18%reduction in client churnPaperchase · AI Engineer

$10Binstitutional portfolio simulatedBMLL Technologies × Columbia

40%fewer manual forecast adjustmentsPaperchase · AI Engineer

50KBloomberg data fields analysed for costJain Global · Quant Developer Intern

20%better forecast accuracy (MAPE)Paperchase · AI Engineer

30%wider anomaly-detection coveragePaperchase · AI Engineer

1stAI adoption strategy case competitionColumbia University · 2026

#85in the US, IMC Prosperity 4Trading competition · 2026

01Featured builds

Finance & Trading

AI at the Margin

A reproducible pipeline that asks twelve language models for stock portfolios, resolves every ticker to an issuer with WRDS data and measures each book on a Barra-style factor model.

12
OpenAI and Anthropic models compared: 2,640
planned model calls across three stages: 7
factor exposures measured for every portfolio: 12
dashboard pages

Read the case study Source

02More work

Systems that ship, not just notebooks

Trading research tools, agentic analytics, forecasting platforms and applied ML. Each one has a full write-up.

All 20 projects

Finance & Trading · Source

AI Adoption Value Realization Dashboard

An executive Power BI system that turns a two-year, $4M enterprise AI adoption strategy into a tracked value-realization layer, modeling spend, revenue uplift, payback, KPI gates, roadmap progress, scenario NPV, and program risk for a $75M CPG e-commerce business.

Power BIDAXTMDL Semantic ModelingFinancial ModelingSynthetic Data Generation

Video demo

Agents & LLMs · Source

MatchOdds AI

An agentic RAG system for NBA pre-game betting analysis that aggregates statistics, injuries, odds, historical matchups, news, and sentiment signals, then compares CoT, ReAct, and multi-agent reasoning strategies for calibrated game prediction.

Agentic RAGSports AnalyticsLLM Agents

Video demo

Systems · Source

Dynamic Resource Allocation in Cloud Computing

A power-aware, SLA-bound deep reinforcement learning scheduler for cloud computing, built with a custom Gymnasium simulator, PyTorch RL agents, FastAPI orchestration, and a React monitoring dashboard.

Deep Reinforcement LearningCloud SchedulingGymnasium

Research · Slides

Data Valuation for Sparse Surface Ocean Carbon Data

A Shapley-based data valuation pipeline for sparse surface ocean carbon prediction, ranking climate and Earth system data sources by their marginal contribution to regional fCO₂ prediction performance.

Data ValuationGeospatial MLOcean Carbon ModelingPandasNumPy

Live demo

Finance & Trading

StressTester

An end-to-end Next.js application that combines deterministic financial simulations with a 4-stage AI pipeline to extract assumptions, generate stress scenarios, compute KPI deltas, and produce mitigation playbooks for enterprise finance teams.

Next.jsTypeScriptLLM Pipelines

Agents & LLMs

Agentic NLP-to-SQL

A security-first, multi-agent natural language to SQL system with clarification, schema grounding, accounting-aware query generation, and safe execution across SQL backends

LLM AgentsNLP-to-SQLSQL Validation & Governance

Finance & Trading · Source

Statistical Arbitrage

A multi-factor statistical arbitrage framework using PCA-based factor models, residual mean reversion, and systematic trade analytics

Quantitative FinanceStatistical ArbitrageFactor Models

03Live lab

Don’t take my word for it. Run it, or watch it.

Deployed apps launch inside the page, and the status next to each is a real health check. The others play a recorded demo.

Open the live lab

04Research

Papers and research

Papers from my research years, mostly federated learning and applied ML. Open a card for the summary and results.

All papers

05Experience

Where I’ve worked

Full resume

August 2026 - Present

Data Scientist – Quantitative Research (Part Time) · Man Group × Columbia - Capstone Project

Engineered a 3-stage LLM experimentation framework evaluating 12 OpenAI and Anthropic models on portfolio construction, automating inference, structured extraction, model evaluation, and cost monitoring.

August 2026 - Present

Data Scientist – Quantitative Development (Part Time) · BMLL Technologies × Columbia - Capstone Project

Built a $10B institutional portfolio simulation using CRSP data, processing 8,452 synthetic orders across 1,010 NAV observations, with daily NAV reconciliation; achieved 107% cumulative pre-cost returns, 16% annualized volatility, and 19% maximum drawdown.

August 2026 - Present

Data Scientist – Quantitative Research (Part Time) · Wedmont Private Capital × Columbia - Capstone Project

Built a Numba lifecycle simulator and a FastAPI service with 25 /v1 operations, addressing every result by a SHA-256 run ID and covering the engine with 27 backend tests.

June 2026 - August 2026

Quantitative Developer Intern – Data Management · Jain Global, LLC

Engineered Bloomberg Data License usage analytics across 50K fields using anomaly detection, cost-impact scoring, and 4-month simulations to identify duplicate, stale, and over-frequent requests and generate field-level optimization recommendations.

January 2026 - May 2026

Researcher · Data Science Institute, Columbia University

Built a data-valuation framework across 13 heterogeneous data sources, training XGBoost models over 8,192 source coalitions and computing exact Shapley values to quantify each source’s marginal contribution to ocean CO₂ prediction accuracy.

April 2024 - July 2025

AI Engineer · Paperchase Accountancy Ltd.

Developed and validated forecasting pipelines using SARIMA, LSTM, N-BEATS, and AutoTS for revenue and demand planning, improving forecast accuracy by 20% MAPE and reducing manual forecast adjustments by 40%.

06Writing