Adithya Phani Thota

VeriRAG, a self-auditing research agent

A RAG agent that checks its own answers. It splits its output into claims, looks for evidence in the retrieved sources, and flags or drops claims it can’t support. The goal is to cut hallucinations, and it costs nothing to run.Self-verification is the main weakness of ordinary RAG, so this has a clear story for interviews.

Brewmate, a multi-agent café app

I had built this project in Google Clouder Builders Pop UP. A multi-agent system for café ordering covering food and drinks. While the customer waits, it generates a personal 15-second AI clip (the “Taste Reel”). The design constraint is that one order plus one clip must cost about $1 of cloud credit.It combines multi-agent orchestration, generative video and cost-aware design.

BillShield, a No Surprises Act medical bill checker

A tool that checks medical bills against No Surprises Act rules and flags likely violations. It was submitted under #social-good.

CareRoute

CareRoute AI is a lightweight healthcare guidance assistant built for the Google Antigravity challenge.

Email Triage Environment

An OpenEnv RL environment where an AI agent triages customer support emails.

Fan Out Lite

AI Ad Localization Agent 6th place, Google DeepMind Hackathon. A localized ad engine. Upload a product and a brief once; the system produces a finished, localized, narrated video ad for each of up to 36 cities across 18 countries in a single chained run — concept, storyboards, per-scene animation, regional scoring, voiceover and muxing, all locked to one shared scene bible. Built for the DeepMind Multimodal Creative Pipelines with GenMedia track.

AI Hackathon Judge Project

The idea is an AI-assisted judging layer that evaluates or scores hackathon submissions, so judging across multiple branch events can be run consistently.