Full-stack work across AI, data, and product

Architected a multi-agent system that connects Outplay's lead research, sequencing, email, and scheduling modules into one autonomous SDR assistant.

• Mapped end-to-end workflows so an AI agent can progress prospects from research to booked meetings.
• Defined dual autopilot and co-pilot modes to balance automation with human approvals.

Built an AI calling agent that verifies lead contact data across telephony providers, keeping sales pipelines accurate and responsive.

• Reduced verification costs 3× and accelerated processing 5× by refactoring call orchestration.
• Lifted contact validation accuracy from 88% to 94% through error log analytics and architecture cleanup.

Shipped an LLM email composer that tailors tone, sequencing, and data cues for every SDR persona.

• Captured product intel, objections, and personas to personalize outbound email content.
• Boosted positive reply rates by 80% across 40 pilot customers.

Delivered a copilot that classifies objections in sales conversations and surfaces context-aware rebuttals.

• Trained intent and NER models that tag objections with 95% accuracy.
• Linked rebuttal playbooks to deal stage, persona, and sentiment cues.

Created a generative engine that assembles personalized multi-channel outreach sequences in minutes instead of hours.

• Cut sequence build time by 90% with persona-aware prompts and context ingestion.
• Streamed each step to the UI as soon as it was ready so SDRs could review progressively.

Modeled ADHD diagnosis indicators with fairness-aware pipelines that keep clinicians in the loop.

• Engineered features from psychometric surveys, cognitive scores, and behavioral logs.
• Benchmarked logistic regression, kNN, and XGBoost with explainability overlays clinicians trust.

Classified muscle signals using ensemble models to power responsive assistive wearables.

• Processed raw biosignals with frequency-domain filtering and window engineering.
• Benchmarked ensemble models, landing on XGBoost with 82.6% accuracy.

Adapted machine learning algorithms to classify galaxy types based on shape and roatation

• Trained multinomial logistic regression model to reach 88.7% macro accuracy.
• Manually created the multinomial logistic regression model using logistic regression base models.

Merged disparate phishing datasets and built machine learning models for real-time URL classification.

• Engineered 30+ URL-based features by synthesizing two public datasets.
• Trained Random Forest and KNN models achieving 97% and 92% accuracy respectively.

Built a Streamlit platform that guides beginners from raw datasets to trained models with side-by-side insight panels.

• Launched a guided workflow covering upload, cleaning, visualization, modeling, and export.
• Implemented dual-panel previews so users see how preprocessing changes affect plots and tables.

Engineered an AI-powered analysis suite that scores sales calls, pinpoints objections, and surfaces next steps.

• Achieved 100% speaker identification accuracy using SpeechBrain embeddings.
• Built a call scoring model spanning tone, action items, and objection handling metrics.

Researched and prototyped automation that records Zoom calls and streams highlights without manual effort.

• Evaluated Zoom SDK pathways to automate recording and live streaming reliably.
• Developed bots that join meetings, capture audio/video, and push via WebSockets for live analytics.

Delivered a suite of IoT prototypes that monitor environments, detect anomalies, and train peers on hardware best practices.

• Built environmental monitors with Arduino and ESP32 nodes reporting to cloud dashboards.
• Developed predictive maintenance alerts that flag anomalies before downtime.