TL;DR:Defines how LLM agents split reconnaissance from purchase, driving the Cost of Information toward zero and undermining dynamic pricing.
Key Insight:A DR-RL pricing policy with DRO ambiguity sets and behavioral KL signals remains robust against agent-contaminated demand.
TL;DR:Built an end-to-end platform that ingests live Belimo actuator telemetry, runs continuous fault classification, and surfaces actionable diagnostics through a facility dashboard and operations agent.
Key Insight:Combining heuristic detection, optional ML inference, and hierarchy-wide fault propagation turns noisy actuator signals into clear system-level health intelligence.
TL;DR:Dynamically places and reevaluates long-running ML training jobs across cloud providers and geographic regions to minimize carbon footprint and cost.
Key Insight:A deterministic score combining carbon-intensity delta and normalized cost drives placement; live reevaluation at set intervals triggers seamless migration when a greener or cheaper node exceeds a configurable threshold.
TL;DR:Built a secure, self-hosted pipeline to extract and cross-reference Spanish notary deeds and tax forms for property transactions.
Key Insight:Automated validation and discrepancy detection reduced manual workload for public officials while keeping data local.
TL;DR:Trained a Seq2Seq LSTM to map anonymized energy bid sequences to market codes, reaching about 89% accuracy.
Key Insight:Chronological splitting, packed sequences, and teacher forcing delivered stable sequence labeling at scale.
TL;DR:Designed and implemented GRAP, a C-based interpreted language with graph-structured syntax, memory, and IO modules.
Key Insight:Graph-shaped programs make data and control flow intuitive without a compilation step.
TL;DR:Built prompt stress-testing and anti-scraping tools and investigated LLM browser exfiltration risks.
Key Insight:Allow-listing inputs and content obfuscation reduce prompt attacks, while red-team work exposed browser data leaks.