Skip to main content

AI Review Assistant

What is the AI Review Assistant?​

The AI Review Assistant is a conversational interface that helps state auditors analyze Medi-Cal Administrative Activities (MAA) cost reports. You ask questions about a report in plain language, and the assistant provides concise, evidence-based analysis with specific numbers, names, and dollar amounts.

Key capabilities​

  • Natural language questions: Ask about fringe variances, salary changes, or any aspect of the cost report data
  • Precomputed report analysis: The assistant uses analysis generated after upload when available, then runs targeted checks only when more evidence is needed
  • Multi-turn conversations: Build on previous questions within the same session -- the assistant remembers what you've discussed
  • Actionable responses: Get specific guidance on what to do, not just what the data shows

Features​

Current Feature Set for sdac-coordinator-release

Position-first review​

  • Prioritizes concrete positions, occupants, and dollar amounts over abstract category summaries
  • Answers questions such as which positions drive a variance, which rows need follow-up first, and where the largest claimable costs sit
  • Best when the reviewer needs a fast triage pass rather than a full audit workflow

Fringe analysis​

  • Explains fringe variances and rate issues in plain language
  • Surfaces likely drivers such as new hires, linked occupants, or unusual fringe amounts
  • Works well for questions like "What is the fringe variance?" and "Which positions are pushing fringe out of range?"

Quarter comparison​

  • Compares the current report with prior-period data when enough context is available
  • Highlights material changes that need justification instead of listing every small movement
  • Can start from district context when a report has not already been uploaded

Replacement cross-reference​

  • Checks replacement rows against the main roster and related comments
  • Helps identify missing links, weak explanations, and occupancy relationships that need district clarification
  • Useful when a reviewer is validating staffing changes or trying to understand a replacement-heavy report

Report summary and prioritization​

  • Produces concise summaries of the top findings in a report
  • Focuses on what should be reviewed next, not just what exists in the data
  • Good for first-pass triage before deeper category review

Conversation memory​

  • Maintains context across follow-up questions in the same conversation
  • Lets the reviewer narrow from a broad question into specific positions or issues without repeating the full setup
  • Reuses persisted conversation state through the chat session model documented in Chat Sessions

Current limits​

  • Optimized for responsive widget use, not exhaustive multi-category review
  • Better at focused questions and fast prioritization than at producing full sendback packages
  • For comprehensive category-by-category review, use Full Review Coordinator

How it works​

You start a conversation by asking a question about a specific cost report. The assistant retrieves the report context and uses precomputed report analysis when it is available. If the stored evidence is missing, still running, or stale, the assistant can run a targeted check and then provide a concise answer. You can ask follow-up questions to dig deeper -- the assistant maintains context throughout the conversation.

Communication style​

The assistant is designed for professional auditors who need direct, evidence-based analysis:

  • Concise -- Gets to the point without lengthy explanations
  • Actionable -- Tells you what to do, not just what is
  • Specific -- Uses exact numbers, names, and dollar amounts (e.g., "Williams added $2,800" not "new hires increased costs")
  • Direct -- Avoids hedging language like "appears" or "seems"

Example interaction​

You: "What is the fringe variance for this report?"

Assistant: "Fringe is 8.7% above threshold. Two new hires (Williams, Lee) account for $5,300. Ask district to update justification."

Use cases​

  • Quick data check: "How many personnel are in this report?" -- Get an immediate count without navigating spreadsheets
  • Variance analysis: "Why did salaries increase this quarter?" -- Understand the drivers behind cost changes
  • Compliance review: "Are there any invalid source codes?" -- Run targeted validation and get plain-language results
  • Sendback preparation: "Summarize the findings for this report" -- Generate a structured correction summary

Session management​

  • Conversations persist across page refreshes -- you can close the browser and come back
  • Sessions automatically expire after 90 days
  • Each conversation is tied to a specific cost report

Next steps​

Chat API

API reference for the streaming chat endpoint

SDAC Overview

Back to the SDAC system overview