MD AMIR KHAN

DXT Commodities

Mar 2026 – Present  ·  Full-time  ·  Stamford, CT (Hybrid)

Agentic AI Engineer

• Designed and built an Agentic RAG system using LangChain and LangGraph to parse unstructured pipeline maintenance notices across 6 operator portal formats — architecting the full pipeline from document ingestion and chunking strategy through embedding model selection, vector store indexing, hybrid retrieval (dense + BM25), and LLM-based structured field extraction
• Engineered prompt templates and structured output schemas for capacity impact extraction — iterating on chain-of-thought prompting and output parsers to reliably extract operator, affected capacity (MMcf/d), duration, and constraint type from free-text notices with varying formats
• Implemented hybrid retrieval with contextual compression (dense embeddings + BM25 ensemble) and MMR re-ranking over a vector store of historical pipeline notices, improving extraction accuracy on ambiguous capacity constraint language by reducing irrelevant context passed to the LLM
• Built evaluation harness using RAGAS-style metrics (faithfulness, answer relevancy, context precision) to benchmark RAG pipeline quality across notice types and operator formats, enabling systematic prompt and retrieval tuning
• Deployed full-stack AI pipeline on AWS EC2 — FastAPI backend exposing RAG extraction endpoints, SQL Server persistence layer for parsed notices, and Power BI dashboard for trading desk consumption — covering 23 interstate pipeline operators

Quantitative Analyst – Market Fundamentals (LNG & Power)

• Designed and built a Permian Basin gas market intelligence system modeling daily production (~22–25 Bcf/d), pipeline egress capacity across 4 major interstate pipelines, and supply-demand balance to predict Waha basis pricing — directly supporting trading desk decisions
• Developed a real-time production estimation model using Genscape pipeline scrape data and EIA 914 reports, training a scaling model (OLS/Ridge regression) with R² and MAE validation to bridge the 2-month EIA reporting lag
• Built a US LNG feed gas forecasting pipeline covering 126 active export trains across 13 terminals (~22,600 MMcf/d nameplate), with 3-model validation framework achieving <3% MAPE on 8-month out-of-sample holdout
• Built an automated scraping and alerting system for EPNG critical notices — polling Kinder Morgan's EBB portal every 5 minutes, deduplicating notices, and pushing real-time Force Majeure and maintenance alerts to Microsoft Teams via webhook, enabling the trading desk to react within minutes of posting
• Engineered end-to-end quantitative pipelines in Python and SQL — from raw data ingestion (Genscape, EIA, FERC bulletin boards) through statistical modeling, cross-validation, and automated forecast output — replacing manual workflows across production estimation, capacity tracking, and pricing analysis

Stevens Institute of Technology

Jan 2024 – Present  ·  Part-time  ·  Hoboken, NJ (Hybrid)

Quantitative Portfolio Research Analyst

• Contributing to research on Robust PCA for dynamic factor portfolio updates — improving stability of risk factors and cutting rebalancing costs by 15%
• Implemented Python-based simulations using corporate bond spread data to study eigenvalue grouping, factor smoothness, and covariance structure robustness
• Developed algorithms for tracking dynamic factors and detecting data structure ruptures to enhance portfolio timing decisions
• Developed and validated ML models (Ridge regression, LightGBM, neural networks) for predicting short-term auction price movements, option pricing, and credit default probabilities
• Researched and implemented advanced risk models including VaR, CVaR, stress testing, and credit risk modeling for MBS, corporate loans, and structured products — covering PD, LGD, and EAD
• Built and analyzed models for mortgage and loan portfolio performance, incorporating prepayment risk, duration/convexity analysis, and sensitivity to macroeconomic variables
• Integrated risk parity, covariance shrinkage, and multi-asset volatility estimation techniques to enhance portfolio resilience and improve Sharpe ratios under varying market regimes