B2U DISTRIBUTED POWER SYSTEM · CALIFORNIA + TEXAS · 75 MWh UNDER BID-ENGINE CONTROL
PHILIP
FELIZARTAPF-00
MARKET OPERATIONS LEAD/B2U STORAGE SOLUTIONS
ENERGY DUCK CURVE DAM × RTM · ARBITRAGE
Philip ships production AI systems end to end. He designed and built the B2U Bid Engine, an autonomous trading bot that operates grid-scale storage built from repurposed EV batteries in California's energy market, and is scaling into Texas. The system runs live every day with millions of dollars in financial consequences. He owns the full ML stack around it: real-time price forecasting, multi-year degradation modeling on the battery packs, and quantitative risk on the portfolio.
In parallel, he is technical founder and CEO of a stealth AI company whose agents run roughly 30,000 autonomous B2B sales conversations a month.
Background: Applied Math and CS at UC Merced, MS in Applied AI at USD. Before B2U, he trained transformers for speech and audio enhancement at Meets The Eye Studios.
SECTION 01 FIELD RECORD
EXPERIENCE.
EX-LOG 2022 - PRESENT
ACTIVE AUG 2023 - PRESENT
LANCASTER, CA
Market Operations Lead
B2U Storage Solutions
↑ PROMOTED JUL 2026from Lead Machine Learning Engineer · Jan 2025 to Jul 2026
↑ PROMOTED JAN 2025from Machine Learning Consultant · Aug 2023 to Jan 2025
Lead developer of B2U's Bid Engine, an AI-driven trading system managing 50 MWh in California and scaling to 100 MWh in Texas. Builds core ML pipelines for asset degradation, incident detection, and financial risk modeling.
Leading agentification of B2U: building the internal AI agent layer to take over the research, monitoring, and reporting work the team currently does by hand.
Own Spark pipeline development in Microsoft Fabric to maintain large-scale battery and market datasets.
Manage frontend tooling for BI and reporting across Power BI, Streamlit, and React; contribute to daily data workflows across market, battery, and site operations.
Founded and lead a stealth AI company building autonomous agents for B2B customer acquisition: agents find prospects and hold high-intent sales conversations, roughly 30,000 conversations per month across a fleet of concurrent browser sessions.
Designed the conversation planner as Monte Carlo Tree Search over LLM sub-agents, so the system plans multi-turn actions under uncertainty instead of following scripted flows.
Built the evaluation loop and the live customer dashboard in React and TypeScript. Leads all sales: closing B2B deals with non-technical buyers and personally running onboarding and campaign reviews.
Built a convolutional + transformer-based chess engine trained with custom self-play data generation. Used Bayesian active learning for hyperparameter tuning across multiple MCTS configurations.
Implemented a max-entropy Nash equilibrium optimizer to construct meta-policies over agents during training.
React + TypeScript app for options analytics: macro “Due Diligence” dashboard with historical distributions, interactive histograms, and options chains. Custom React visualizations for multimodal price distributions.
Inverted European call/put prices into implied discrete probability distributions via a discretize-then-optimize framework. LLM integration converts written market views into quantitative shifts of the implied distribution.
Applied L-BFGS on model-independent pricing equations; tested on VIX/SPX to highlight crashophobic sentiment. Built omega-ratio constrained optimizer modeling asset dependencies with empirical multivariate copulas.
