Subash Matu

• Engineered automated diagnostic tools, including a Python/Playwright web scraper and a Twilio API/PostgreSQL-backed automated test caller, to navigate complex phone trees and verify ILS number routing (e.g., Zillow, Apartments.com). Successfully diagnosed root causes of low inbound traffic and eliminated infinite call loops to prevent dropped calls and improve user experience.
• Conducted deep root cause analysis using Snowflake, Datadog, Hex, Salesforce, by building diagnostic pipelines with conversation and sentiment analyzers. Curated key performance metrics (dial-loop rates, lead conversion, abandonment) to directly support cross-functional teams during high-stakes client outreach and customer support conversations.
• Analyzed and optimized call routing strategies across an initial 8,000+ properties, scaling impact to tens of thousands of buildings; identified a 5.6x difference in abandonment rates to drastically improve phone tree configurations.
• Spearheaded live, client-facing discussions to save major enterprise accounts, presenting structured VoiceAI performance audits and deep-dive analytics. Leveraged Zendesk trend data in real-time to prove product ROI and clearly separate AI limitations from client operational gaps.
• Architected a fully automated, end-to-end EIN collection and Twilio A2P 10DLC compliance lifecycle using Hex, Zapier, Snowflake, and Slack APIs. Integrated Customer.io and Claude Cowork to trigger CSM-approved client outreach via Slack reactions, tracking profile updates dynamically until final Twilio approval, eliminating manual onboarding bottlenecks.
• Created comprehensive QA documentation and interactive visual architecture maps (Miro) to demystify complex VoiceAI data flows, successfully utilizing these resources to onboard and train multiple cross-functional team members.

• Architected a privacy-centric, dual-audience AI RAG system that securely processes patient data to deliver real-time diagnostic analysis for physicians while simultaneously generating accessible, demystified hospital stay summaries for patients.
• Developed an end-to-end Python pipeline for SEEG signal reconstruction by fine-tuning a CAE-LSTM model (achieving a Pearson correlation of 0.9) and engineered a robust evaluation framework using statistical similarity metrics (Hurst exponent: 0.72, Mutual Information: 1.34) to significantly improve brain signal interpretability for clinical applications.

Worked with large-scale datasets from multiple banks, developing automated pipelines and data cleaning protocols to extract, analyze, and present actionable financial insights to clients and stakeholders.

Developed a hybrid ML model combining EfficientNet and triplet networks to detect signature forgeries with 87% accuracy. Leveraged advanced preprocessing and image augmentation techniques to enhance model robustness. Presented model insights to stakeholders through visualizations, explaining fraud detection performance and business impact.

Led and built a Website Development team to create and deploy a full-stack website, integrating a data analytics module that plotted user interaction data and refined inventory management, significantly boosting efficiency

I spearheaded the design and development of a fully functional website, incorporating features like shipment tracking, dynamic fluid animations, and contemporary design principles. Additionally, I implemented a user-friendly CMS portal for seamless content management and collaborated closely with cross-functional teams to integrate diverse functionalities.

Engineered a data pipeline to ingest over 100,000 betting records from Polymarket and the Polygon Blockchain, capturing individual-level betting behaviors for the 2024 USA election and analyze cross-market betting behaviors and patterns. Applied a Bayesian cascade model (inspired by the monotone likelihood ratio property) to quantify herd behavior and the “wisdom-of-crowd” effects, modeling each bettor’s decision as a signal that updates the posterior probability of adoption. Uncovered a higher correlation between early adopters’ signals and eventual market trends, highlighting fragile cascades and conditions under which later bettors deviate from early signals.

Developed a subject-specific SEEG signal reconstruction pipeline by fine-tuning a pre-trained CAE-LSTM model, leveraging advanced preprocessing and optimizing functional connectivity analysis (Granger causality, PLV) to model neural interactions, improving reconstruction accuracy and achieving a Pearson correlation of 0.9. Engineered a robust evaluation framework for SEEG signal reconstruction, integrating statistical similarity metrics (Hurst exponent: 0.72, Mutual Information: 1.34) and dynamic visualizations. Enabled deeper insights into brain signal reconstruction, improving interpretability for neurological research and clinical applications.