Anshuman Mishra Email & Phone Number

At Brainchain, we use LLMs at scale to identify and exploit supply chain disruptions.In total, the systems I developed have generated opportunities with over $200MM in associated revenue over the last 12 months.* As the first engineering hire, I built out Brainchain's event detection pipeline, consuming approximately 1.5 million news data points per day to identify supply shocks, demand spikes, and events that may affect downstream products and commodities. * Since this was at the dawn of LLMs as a service, we had to solve a ton of LLMOps functionality in house, including: - progressive summarization, - long-term memory, - RAG via HyDE,- load balancing,- structured data extraction via schema coercion, and- automated knowledge graph construction and enrichment* I designed and built Brainchain's business plan creator, using LLMs in parallel to generate fully fleshed-out B-plans with links to relevant rules and regulatory requirements, integrating search via function calling.* I designed a TAM sizing engine that would use multiple simulated scenarios and perform step-by-step reasoning to create a weighted estimation of total addressable markets for a given opportunity. * Built an engine to analyze recent laws passed by the US Congress, including the Inflation Reduction Act, the Defense Appropriations Act, and the CHIPS Act, seeking opportunities marked out as pork barrel spending.

At Numenta, my work involves developing novel techniques to make large language models (eg BERT and friends) faster to train, faster at inference, and smaller, while reducing ML model running costs and making excellent accuracy-speed/size tradeoffs for enterprises or startups.Language models are changing the world - they can compose poetry, write your school term paper (don't do that), and come up with website copy. However, they are resource-hungry, requiring lots of compute power and memory to be able to do these things.I focus on using speedup techniques from multiple areas (linear algebra, optimization, compact data structures, randomized algorithms, etc) and combining them with classic model optimization (matrix factorizations, knowledge distillation, pruning, etc) in order to better exploit neural network dynamics at both training and inference time.

I built an engine to perform automated stock picking for the US markets using deep learning (for SEC EDGAR filings and news), machine learning (on features extracted by blending fundamental, macroeconomic, and price/volume-based technical data), and reinforcement learning (for the overall stock picking system). The engine analyzed 4000+ US stocks, sectors, industries, interest rate curves, commodity markets, corporate activities (earnings, M&A announcements) and economic news on a daily basis, deriving 1000+ features per stock, and ranking those names that were likely to outperform the market over the near term.An article I wrote on the engine architecture went viral on Hacker News in 2022, garnering ~200k views: https://principiamundi.com/posts/didact-anatomy/

Santa Clara, California, Us

Built from scratch the core platform for TIBCO’s Automated Machine Learning project, with end-to-end ML pipeline generation from automated feature engineering (powered by deep feature synthesis on rich cross-linked DB/warehouse tables) to model selection and tuning via hyperparameter optimization. Created custom functionality to perform Monte Carlo simulation experiments at scale using Spark.

Santa Clara, California, Us

Santa Clara, California, Us

Customer-focused data science projects for clients in Financial Services and Pharma. Built solutions in Scala, Python on Apache Spark for “Next Best Action” recommender systems, cybersecurity, intrusion detection, derivatives reconciliation, and fraud detection. Spoke at the Spark Summit in 2016 on “Enterprise Scale Topological Data Analysis (TDA) using Spark”, demonstrating the use of locality-sensitive hashing techniques to boost performance of TDA by ~10x on Spark clusters.

Designed, implemented and soft-launched the Ayasdi Market Signals Service (directional forecasts data feed for S&P 500 and 9 equity sectors). Identified and pitched prospects in the US hedge fund industry, built and demonstrated rigorous statistical back-tests, led technical discussions for 10+ trials with major East Coast-based hedge funds.Published a US patent: Topological data analysis for identification of market regimes for prediction

Natick, Ma, Us

Responsible for product management of the Computational Finance product family at MathWorks. Subject matter expert, product owner of Database Toolbox, Financial Instruments Toolbox, and Trading Toolbox.Built a MATLAB-based trading agent that performs a time-series similarity search on ~44,000 days of 1-minute data in <500 milliseconds to predict and trade on market moves for NASDAQ futures.Wrote paper on cross-sectional momentum trading via quant modeling: “Catching up with the trend”, shortlisted for 2014 Wagner Award by National Association of Active Investment Managers (NAAIM)

New York, Ny, Us

Designed systematic strategies and made markets in G10 FX swaps.Traded daily O/N, T/N balances for G10 FX in the interbank market. Developed new FX swaps trading strategy by statistical modeling of swap price using desk inventory position, per-trader and desk risk limits, demand/supply forecasting, and market conditions as features.

New York, Ny, Us

1. Equities – Structured Volatility and Exotic Options Trading. Analyzed structured note term sheets to estimate desk risk exposure, created an Excel/VBA application for accurately pricing a merger arbitrage-based structured note.2. Structured Capital Markets. Researched tax arbitrage strategies in corporate settings. Designed high-frequency trading strategies for the AUD/NZD and EUR/CHF currency pairs.3. Quantitative Prime Brokerage Product Management. Developed an automated reporting tool to analyze equity volumes across multiple dark pools and execution venues such as NYSE, NASDAQ, BATS.

Gurgaon, Haryana, In

Traded equity index futures: E-Mini S&P 500 and Eurostoxx 50.* Trading strategy design for spot/futures markets in equities, equity indices, FX, and commodities using:a. statistical modeling techniques - machine learning and predictive modeling, anomaly/outlier detection, motif recognition, ensemble methods, time series classification and similarity search techniques for market context injection, regime shift and crash forecast modeling using LPPL modelsb. quantitative and classical factor-driven models - momentum, carry, term structure, volatilityc. data mining-based models - sentiment analysis, now-casting for nascent trends, creating ensembles for economic regime shift detectiond. exploitation of market micro-structure and order book information to inform order type and sequencing* Statistically rigorous strategy testing methodologies to detect presence and decay of alpha, including random-start back-testing, block bootstrap back-testing, Monte Carlo simulation-based analysis, walk-forward analysis* Real-time market indicators to measure/proxy liquidity, instantaneous variance, short-term mean reversion * Tactical asset allocation and portfolio construction using higher-order moments, risk parity, enhanced risk parity, minimum correlation, and projected tail-risk based allocation strategies* Designed medium-frequency trading strategies for futures based on unsupervised statistical learning techniques. * Designed mediurm/high-frequency intraday trading strategies based on statistical analysis and time series modeling of the spread between European and American equity index futures, taking into account weekday and hour-of-day seasonality effects.

Ivontu.com was a startup working in the field of machine-directed sentiment analytics for financial news, enabling the user to create and train models for specific stocks.Designed a machine learning framework to determine the probabilistic effects of incoming news on stocks in the S&P500. Implemented the design in Java and rolled out a custom front-end AJAX library to power the site.Consulted with Indian firms to license the framework for use in econometric analysis of Indian commodity markets.

Co-founder of a hot Bangalore-based startup. Managed the production of India’s first social networking-based call filtering service. Identified target segments across Indian cities and created effective go-to market strategies. Led media outreach efforts, and was interviewed by national and international media such as NYTimes/International Herald Tribune and LiveMint (WSJ India).

Austin, Texas, Us

Languages - Java (Java EE frameworks), C, Javascript

Glasgow, Glasgow, Gb

As a summer intern in the Department of Computer Science, I worked on the design and implementation of a pluggable typechecker generating system for the SableCC parser generator.

Bachelor Of Technology (B.Tech.), Computer Science

Master Of Technology (M.Tech.), Computer Science

Mba, Investment Management, Finance

Isc

Cfa Level I In June 2007, Cfa Level Ii In June 2008

Cleared Frm In December 2007

Education record