Valarix Agent: The AI Tax Auditor
Inspiration As a CPA, I saw automation solve compliance but fail at communication. Clients still waste time asking "how do I fill this?" during video calls. Valarix Agent fixes the "last mile" of accounting by providing real-time, spoken guidance during the actual filing process.
What it does An autonomous AI participant for Google Meet that:
Listens to transcripts via Recall.ai.
Watches shared screens to "see" tax forms via Gemini 2.0 Flash.
Speaks answers directly into the call using Google TTS when triggered by voice questions.
The Tech Stack Frontend: Lovable (React/Tailwind).
Hosting: Render (Application) & Supabase (Database/Edge Functions).
Core: Antigravity for backend logic.
AI: Gemini 2.0 Flash (Vision/Reasoning) & Google Cloud TTS (Voice).
Challenges & Lessons Multi-Modal Orchestration: Combining transcription, LLM reasoning, and voice generation into a single "human-like" flow is incredibly difficult. We learned to handle multiple API integrations concurrently to minimize latency.
The Scalability Gap: Moving from a single-user prototype to a multi-tenant system requires heavy orchestration. We utilized Render to stabilize the production environment.
Bridge Audio: We learned that "thinking" time kills the experience. We implemented immediate verbal acknowledgments to bridge the gap while the AI analyzes the screen.
Accomplishments Successfully implemented a live support agent that can visually identify a tax document and speak advice in real-time.
What's next Scaling via Render to support concurrent sessions and integrating the agent into valarix.com to launch the first global autonomous accounting protocol.
Repo
Built With
- antigravity
- gemini
- lovable
- render
- you.com

Log in or sign up for Devpost to join the conversation.