MD AMIR KHAN
AI Engineer & Quantitative Analyst
AI engineer and quantitative analyst at DXT Commodities, building production systems for U.S. natural gas and power markets — LLM pipelines, forecasting models, and full-stack analytics platforms used daily by the trading desk.
About Me
I'm an AI engineer and quantitative analyst at DXT Commodities (Stamford, CT), building production systems for U.S. natural gas and power markets — the trading and analytics teams use them daily.
On the AI engineering side: an LLM pipeline that reads unstructured maintenance and capacity notices across ~20 U.S. interstate gas pipelines, a real-time Force Majeure alerting service covering 30+ interstate pipelines that pushes notices to Microsoft Teams within minutes of posting, and a full-stack analytics platform covering the PJM generation fleet at national scale.
On the quant side: a U.S. LNG feed-gas forecasting pipeline achieving sub-3% MAPE on out-of-sample validation, a national gas demand forecaster spanning 12 regions × 4 end-use sectors, a real-time production model that bridges the two-month EIA reporting lag, and an FTR nodal-basis screen on 8+ years of PJM day-ahead settlement data.
Alongside DXT, I supported Prof. Papa Momar Ndiaye's research at Stevens on the ε-subdivision Robust PCA framework for dynamic factor portfolios — implementing the algorithm in Python and validating it on ~6.5 years of daily returns across the 11 S&P 500 GICS sector ETFs. Working paper in preparation.
Core expertise:
- AI Engineering: LLM pipelines, LangChain / LangGraph, RAG, LLM agents, structured extraction, prompt engineering
- Machine Learning: Scikit-learn, XGBoost, Ridge / OLS, time-series forecasting, walk-forward cross-validation
- Full-Stack: Next.js, React, TypeScript, FastAPI, SQL Server, Docker, AWS
- Energy Markets: Natural gas & power fundamentals, LNG feed-gas, pipeline capacity, ISOs / RTOs (PJM, ERCOT, NYISO, MISO, ISO-NE), FTR markets
Education
Master of Science in Financial Engineering & Analytics
Stevens Institute of Technology · Hoboken, NJ, USA
Focus: Quantitative Finance, Algorithmic Trading, Risk Analytics, Portfolio Optimization
Bachelor of Business Administration
North South University · Dhaka, Bangladesh
Major: Finance · Minor: Mathematics
Key Coursework: Calculus, Linear Algebra, Differential Equations, Corporate Finance, Investment Theory, Financial Derivatives, Applied Statistics
Latest News
New Role
Joined DXT Commodities as AI Engineer & Quantitative Analyst
Started full-time at DXT Commodities (Stamford, CT) in March 2026 on the Market Fundamentals team, covering LNG and natural gas markets. Building LLM extraction pipelines, LNG feed-gas forecasting, and real-time trading-desk alerting infrastructure.
In Progress
AI Engineering from Scratch — Open-Source Curriculum
Currently working through AI Engineering from Scratch, Rohit Ghumar's open-source curriculum (20 phases, 503 lessons across Python, TypeScript, Rust, and Julia). It builds AI systems from raw math up — linear algebra and backpropagation through tokenization, attention mechanisms, and autonomous agent systems — implementing each algorithm from first principles before touching frameworks. Committed to completing the full curriculum to strengthen the foundational layer beneath my applied LLM work.
Course Completed
Advanced RAG (Retrieval-Augmented Generation) — May 2026
Completed a 10-module Advanced RAG course from CampusX covering the full LLM pipeline stack — embeddings, vector stores, hybrid retrieval, HyDE, CRAG, Self-RAG, Graph RAG, Agentic RAG with LangGraph, and RAGAS evaluation. Directly applied to production AI systems at DXT Commodities.
Research Paper
Towards a Robust PCA and Dynamic Factor Portfolios Updating
Working paper with Prof. Papa Momar Ndiaye on the ε-subdivision Robust PCA framework for dynamic factor portfolios — block decomposition of the eigenspectrum against a tolerance ε, Gram–Schmidt construction of the closest orthonormal basis to the prior period's factors, and rupture-detection that resets factor tracking when block-mean eigenvalues shift beyond a threshold δ.
Validated on ~6.5 years of daily returns across the 11 S&P 500 GICS sector ETFs, spanning the COVID-19 shock and the post-pandemic inflation cycle. The robust approach stabilized factors at portfolio volatility essentially identical to standard PCA. SSRN preprint in preparation.
Experience
DXT Commodities North America
Mar 2026 – Present · Full-time · Stamford, CT (Hybrid)
Applied AI Engineer & Quantitative Analyst — Market Fundamentals (LNG & Power)
- Pipeline transparency platform. Built the scraping and LLM extraction pipeline that reads maintenance and capacity notices across ~20 U.S. interstate gas pipelines and multiple operator portal architectures. Structured extraction with the Claude API turns free-form notices into typed capacity-impact records. FastAPI + SQL Server.
- PJM Fleet Analytics Platform. Full-stack platform covering the PJM generation fleet at national scale. Overview, thermal, non-thermal, plants, insights, and natural-language query views used daily by the power desk. Next.js + React + TypeScript, FastAPI, SQL Server, Docker.
- Multi-ISO Power Price Analytics Platform. Live platform covering PJM, ERCOT, NYISO, MISO, and ISO-NE. Historical LMPs by energy/congestion/loss component, forward-contract settlements with strip aggregation, and a Price ⇄ Heat-Rate toggle that connects power and gas.
- U.S. LNG feed-gas forecasting. End-to-end forecasting pipeline covering the U.S. LNG export terminal fleet. Three-model validation framework; sub-3% MAPE on out-of-sample validation.
- U.S. natural gas demand forecasts. Multi-model forecaster (12 regions × 4 end-use sectors) producing daily 30-day forecasts for the Lower 48. XGBoost for weather-sensitive sectors, Ridge for the slower ones. Driven by NWS temperature forecasts and Fed industrial production data, trained on EIA state-level consumption; walk-forward cross-validation.
- Real-time gas production model + Permian Basin intelligence. OLS/Ridge scaling framework that combines licensed pipeline-nominations data with EIA monthly statistics to produce a current-month U.S. production estimate inside the EIA reporting lag. Permian codebase covers daily production, egress capacity, and Waha basis pricing.
- Pipeline Force Majeure alerts + PEPCO nodal basis screen. Real-time alerting service polls electronic bulletin boards across 30+ U.S. interstate gas pipelines every 5 minutes and pushes Force Majeure and maintenance notices to Microsoft Teams (response time: hours → minutes). PEPCO FTR-bidding screen covers 8+ years of PJM day-ahead settlement data.
Stevens Institute of Technology
Apr 2025 – Feb 2026 · Hoboken, NJ
Quantitative Research Assistant — School of Business
- Implemented the ε-subdivision Robust PCA framework for dynamic factor portfolios in Python with Prof. Papa Momar Ndiaye — block decomposition of the eigenspectrum against a tolerance ε, Gram–Schmidt construction of the closest orthonormal basis to the prior period's factors, and rupture-detection that resets factor tracking when block-mean eigenvalues shift beyond a threshold δ. Added K-Means clustering on eigenvector features as an independent signal of factor-structure change.
- Validated on ~6.5 years of daily returns across the 11 S&P 500 GICS sector ETFs, spanning the COVID-19 shock and the post-pandemic inflation cycle. Fed the robust and standard covariance estimates into a mean-variance optimizer and compared cumulative returns, 126-day rolling annualized volatility, and cluster-transition timing — the robust approach stabilized factors at portfolio volatility essentially identical to standard PCA.
Projects
Pipeline Transparency Platform (LLM Extraction)
DXT · ProductionLive web application monitoring ~20 U.S. interstate natural-gas pipelines across multiple operator portal architectures. Built format-specific scrapers (HTML, PDF, Excel, dynamic and protected portals) and a structured-output extraction layer over the Anthropic Claude API that normalizes free-form maintenance notices into typed capacity-impact records. Caching layer significantly reduces LLM calls on unchanged notices. Tracks active and upcoming capacity restrictions across the interstate pipeline network.
PJM Fleet Analytics Platform
DXT · ProductionFull-stack production web application covering the PJM generation fleet at national scale — thousands of units across 13 states + DC. Overview / Thermal / Non-Thermal / Plants / Insights / Ask (natural-language query) / My Units / Exports views used daily by the DXT power desk.
Multi-ISO Wholesale Power Price Analytics Platform
DXT · ProductionLive platform for U.S. wholesale power analysis across the five FERC-regulated RTOs (PJM, ERCOT, NYISO, MISO, ISO-NE). Historical daily and hourly LMPs with energy/congestion/marginal-loss components, forward-contract settlements with strip aggregation (Summer, Winter, Spring, Q1–Q4), and an integrated Price ⇄ Heat-Rate toggle bridging power and gas markets via commercial gas indices.
U.S. LNG Feed-Gas Forecasting Pipeline
DXT · ProductionEnd-to-end demand forecasting pipeline covering the U.S. LNG export terminal fleet — the full production terminal set across the Gulf Coast and East Coast. Three-model validation framework achieving sub-3% MAPE on out-of-sample validation.
U.S. Natural Gas Demand Forecast Platform
DXT · ProductionDaily 30-day demand forecasts for the U.S. Lower 48, disaggregated into 12 geographic regions × 4 end-use sectors (specialized machine-learning models per region-sector combination). XGBoost for weather-sensitive sectors, Ridge regression for slower-moving ones. Driven by federal weather and macroeconomic data; trained on EIA state-level consumption. Walk-forward cross-validation with per-fold accuracy metrics and physical-bounds sanity checks.
Real-Time U.S. Natural Gas Production Model
DXT · ProductionOLS / Ridge scaling framework that combines licensed daily pipeline-nominations data with EIA monthly production statistics, validated by R² and MAE. Produces a current-month U.S. natural-gas production estimate that operates within the EIA reporting-lag interval.
Permian Basin Market Intelligence System
DXT · ProductionDaily codebase modeling Permian basin production, integrating egress capacity across the major egress pipelines to construct a supply-demand balance and predict Waha basis pricing for the trading desk.
Pipeline Force Majeure Alert Service
DXT · ProductionReal-time alerting service that polls electronic bulletin boards across 30+ U.S. interstate natural-gas pipelines every five minutes, deduplicates events against persistent state, and pushes Force Majeure and maintenance notices to the DXT trading team via Microsoft Teams webhook. Cuts trading-team response time from hours to minutes.
PEPCO Nodal Basis Screen (FTR Bidding)
DXT · ProductionPhase-1 quantitative screen of PJM bidding nodes in the PEPCO utility zone for downstream Financial Transmission Rights bidding analysis. Millions of hourly observations across dozens of nodes plus the zone aggregate, drawn from 8+ years of PJM day-ahead settlement data. Deliverables: print-ready PDF stakeholder report, per-node per-component CSV table, and a full-panel parquet dataset for reproducibility.
Bond Portfolio Optimization and Immunization
August 2025Comprehensive bond portfolio management system combining quantitative finance with data engineering. Implements duration matching, convexity adjustments, and immunization strategies using real-time data pipelines, automated risk calculations, and scalable portfolio optimization algorithms for fixed income portfolios.
Vasicek Bond Pricing Model - Monte Carlo, PDE & Analytical
July 2025Comprehensive implementation of the Vasicek interest rate model featuring three pricing approaches: analytical solutions, Monte Carlo simulations, and PDE finite difference methods for zero-coupon bonds.
Portfolio Optimization
July 2025Strategic asset allocation framework using modern portfolio theory, risk parity, and advanced optimization techniques with Riskfolio-Lib for multi-asset portfolio construction.
Stock Brokerage System Low Level Design
February 2025High-performance stock brokerage system architecture featuring order matching engine, portfolio management, and real-time market data processing.
Option Pricing Models
February 2025Comprehensive options pricing library implementing Black-Scholes, binomial trees, and Monte Carlo methods for European and American options valuation with Greeks calculation.
SPY Momentum Alpha Backtesting
February 2025High-frequency momentum trading strategy combining data engineering and quantitative finance. Built robust data pipelines processing 2 years of SPY tick data from Polygon API, implemented real-time signal generation, and achieved 79% total return with comprehensive performance analytics and automated backtesting frameworks.
Pairs Trading Strategy
February 2025Statistical arbitrage strategy using cointegration analysis and mean reversion. Employed Euclidean distance method for pair selection with z-score based entry/exit signals.
Options Pricing Using Machine Learning
September 2024Advanced machine learning approach to options pricing combining deep learning with financial engineering. Implemented neural networks, random forests, and ensemble methods with automated feature engineering, model validation pipelines, and real-time pricing systems that outperformed traditional Black-Scholes pricing in complex market conditions.
Activities & Awards
Student Membership
CFA Society New York
Student member, actively engaged in professional events.
Certifications & Licenses
Anthropic Education — Certified Track
Anthropic
April 2026Completed Anthropic's official education program covering the full Claude API and agent-development stack — API fundamentals, Claude Code, the Model Context Protocol (intro + advanced), agent skills, and subagents. Directly applied to production LLM systems at DXT Commodities.
Advanced RAG (Retrieval-Augmented Generation)
CampusX
May 202610-module course covering the full RAG stack — from document processing, embeddings, and vector stores through advanced retrieval techniques (HyDE, CRAG, Self-RAG, Graph RAG), Agentic RAG with LangGraph, and production deployment. Directly applicable to LLM-powered pipelines at DXT Commodities.
AI Engineer Bootcamp 2026: LLMs, RAG, AI Agents & Vector DBs
Udemy · Paulo Dichone
April 202628-hour, 306-lecture bootcamp covering the full applied AI engineering stack — LLMs, retrieval-augmented generation, autonomous AI agents, and vector databases. Direct overlap with the production LLM systems I build at DXT Commodities.
Vector Databases: Fundamentals to Production (2026 Edition)
Udemy · Paulo Dichone
May 2026Applied course covering the vector-database stack end-to-end — embedding models, indexing strategies, similarity search, and production integration into RAG pipelines. Directly supports the retrieval layer in the LLM systems I build at DXT.
Complete Algorithmic Trading Course with Python, ChatGPT, ML
Udemy
July 2025Comprehensive algorithmic trading course covering Python programming, machine learning integration, and ChatGPT applications for automated trading strategies.
Akuna Capital Options 101
Akuna Capital
July 2025Professional options trading course from leading market maker covering payoff diagrams, volatility, Greeks, and market-making fundamentals.
Complete Data Science, Machine Learning, DL NLP Bootcamp
Udemy
July 2025Comprehensive bootcamp covering data science fundamentals, machine learning algorithms, deep learning, and natural language processing applications.
FastAPI - The Complete Course 2025 (Beginner + Advanced)
Udemy
July 2025Modern Python web framework for building high-performance APIs, essential for financial data services and algorithmic trading platforms.
Full-Stack Web Development Track
Udemy · Colt Steele · Stephen Grider · Dr. Angela Yu
May – July 2026Consolidated full-stack curriculum spanning ~220 hours across five courses from Udemy's top instructors — modern JavaScript and TypeScript foundations, React (Hooks, Context, Next.js, Router), Node.js microservices, and end-to-end full-stack web development. Underpins the front-end and API layer of the analytics platforms I build at DXT.
Developer Foundations Track
Udemy · Colt Steele
June – July 2026Foundational developer tooling — the Linux command line (16 hours) and Git & GitHub for team workflows. Underpins day-to-day production engineering practice at DXT.
Probability — The Science of Uncertainty and Data
MIT / edX
December 2022Rigorous probability theory course covering uncertainty quantification, statistical inference, and data analysis fundamentals from MIT.
Python and Statistics for Financial Analysis
Coursera
February 2022Specialized course combining Python programming with statistical methods for financial data analysis and investment decision making.
Technical Skills
Languages & Tools
AI / LLM
Machine Learning
Energy Markets
Resume
MD Amir Khan — Resume
AI engineer & quantitative analyst · U.S. natural gas & power markets. Full detail on my DXT work, Stevens Robust PCA research, education, and technical skills.
Get In Touch
mkhan37@stevens.edu
Location
Stamford, Connecticut, USA
linkedin.com/in/amirkhan2317
Portfolio
Get In Touch
I'd love to hear from you! Please feel free to reach out through any of the following methods:
Email: mkhan37@stevens.edu
Location: Stamford, Connecticut, USA