Proctor
Team consisting of AI and software engineers from Rattle and IT Solutions with MS degrees from ASU, IIIT-H, and Texas Tech, specializing in Rust, Python, AWS Bedrock, and RAG pipelines.
YouTube Video
Project Description
Proctor is a privacy-first AI supervisor that lets medical and insurance professionals use the reasoning power of Gemini without ever exposing sensitive data. Our agentic workflow uses FunctionGemma and Cactus Compute to redact PII locally, whisper for voice transcription and it routes only anonymous corpus to the cloud while keeping the human in total control via human in the loop approval. Cloud model neves sees the form with PII, and the agent intelliently decides weather to route the model to cloud fallback based on the complexity of the query involved.
Technologies & Frameworks
Proctor utilizes a specialized stack to enable dynamic escalation and local-first goals:
1) FunctionGemma: Local SLM for PII discovery and intent routing.
2) Gemini 1.5 Pro: Cloud engine for complex, multi-modal reasoning.
3) Cactus SDK: Powers high-speed, on-device inference at the edge.
4) Cactus - Whisper: Enables 100% local voice-to-text for the voice agent.
5) PyMuPDF, python-docx : Local tools for OCR and document extraction.
6) Gradio - To create a simple ui for the demo.
Prior Work
This project was inspired by the work I have done in my internship; which involved annonmizing data using ineffiecient nlp techniques that create crude regex for common PIH patterns.