1. Building agentic infra and auth, focused on reliability, trust, and verification.
2. Building end-to-end pipelines for AI-native automation, from agent harnesses to biomedical engineering.

Silicon Valley, CA · rithvik1788@gmail.com

• Build custom VoiceOS integrations with real OAuth, no Composio and no third parties: say "Connect [App]" and the provider's own approval page opens, with the token minted straight into the macOS Keychain
• One config file per provider and zero auth code needed, held to 625 tests
• Read the docs, written so an AI agent can install and wire it for you end to end

• A deterministic gate for VoiceOS that runs before an agent fires any irreversible action: every routing parameter (recipient, amount, destination) must trace back to a span of what the user actually said, blocking prompt injection structurally
• Pure string and number matching gives the same verdict for the same input in microseconds, wrapping any MCP tool server in shadow mode until you flip it to enforce
• Read the docs, written so an AI agent can install and wire it for you end to end

• Use Slack by voice through VoiceOS: catch-up digests, read, search, send, schedule, threads, status
• 16 tools, one-tap OAuth on Slack's own approval page, built on the Native OAuth Engine with zero auth code

• Automated end-to-end pipeline for zero-cost detection of aneurysm rupture risk at AUC 0.834 and 80% sensitivity, with a per-patient explanation behind every decision
• Drives CFD through the SimScale API, feeds structural FEA, and trains an interpretable Random Forest with a SHAP explanation engine, replacing commercial workflows that cost $20k to $40k a year and take hours per case with a free pipeline that takes about 24 minutes per aneurysm
• 🥇 1st Award, Biomedical Engineering @ Synopsys Silicon Valley
• 🏅 5th in California, Medicine & Physiology @ CSEF 2026
• 📄 Read the research paper

• Collective intelligence for AI agents: solve a task once, it becomes a skill any agent can reuse
• Skills earn live trust scores from real pass or fail outcomes, savings counted in tokens never generated
• npm for agent knowledge, built with Arav Dharnikota: live demo

• Say "open a whiteboard and explain transformers" and a glass lightboard opens beside your call
• A live AI tutor draws on it while it talks, one continuous performance driven by tool calls
• Built with Arav Dharnikota at the VoiceOS Hackathon, August 2026: live demo, sped up