Naga Pavithra
Lagisetty

Emergency call centers get overwhelmed during crises and critical calls get lost in the queue.

• Built a real-time triage system to handle high-volume emergency calls using AI-driven prioritization (P0–P3)
• Integrated Twilio for telephony and ElevenLabs for conversational voice AI to classify and route critical calls to human agents
• Developed a live monitoring dashboard to track call queues, escalation status, and agent allocation

Small businesses lose revenue when they miss customer calls. Callers hang up and never come back.

• When a customer calls a small business and the call is missed, the AI automatically initiates a chat conversation with the caller via iMessage
• The AI assistant answers questions about the business, handles inquiries, and books an appointment, all without human intervention
• Built with LLMs, Twilio APIs, and async Python. Handles multiple users texting at once with concurrent session management and real-time message streaming

SOC analysts see 10,000+ alerts daily. 67% are noise. Real attacks slip through.

• Built a three-agent pipeline on NVIDIA NemoClaw + Nemotron that triages alerts, verifies its own decisions, and recommends response plans — humans approve before anything executes
• Achieved 76.7% triage accuracy across 30 synthetic SIEM alerts with a self-correcting verifier agent that catches and adjusts misclassifications
• Full pipeline processes each alert in ~8.5s (triage → verify → response plan) using vLLM's high-throughput inference

Startups fail from chain reactions nobody mapped. Risks compound silently across domains.

• Built a multi-agent AI system to identify operational risks and failure cascades in startups before they become catastrophic
• Orchestrated 6 specialist agents (People, Finance, Infrastructure, Product, Legal, Code Audit) running concurrently via LangGraph
• Implemented cross-domain cascade expansion using Gemini 2.5 Pro, producing a 0–100 risk score with a plain-language founder briefing
• Built an interactive React dashboard with real-time SSE streaming and React Flow cascade visualization

Non-technical users struggle to extract insights from data without knowing SQL or Python.

• Built a web-based analytics tool that lets users upload CSV files and analyze them using natural language queries powered by Google Gemini AI
• Developed a multilingual benchmark comparing GPT and Gemini for SQL/Pandas query generation from natural language
• Built a React + TypeScript frontend with Recharts visualizations and a FastAPI backend with pandas/numpy processing

Women walking alone at night lack a discreet, instant way to alert someone if they feel unsafe.

• Built an AI-powered safety companion that engages users in real-time conversation during solo walks via ElevenLabs voice AI + Twilio
• Implemented a custom safe-word system that automatically sends an iMessage with live location to emergency contacts when triggered
• Developed a FastAPI backend with SQLite user registration and ngrok webhook tunneling

Renters have no easy way to assess neighborhood safety when browsing Zillow listings.

• Developed a Chrome extension that evaluates safety scores of Zillow listings using Boston PD crime data
• Built a predictive model using XGBoost with geolocation features to assess neighborhood safety by zipcode and street
• Deployed the model via Streamlit for real-time safety insights