SmartViewer: AI Agent Logic Map

Architecture: RAG + ReAct framework with Entity Resolution
Mission: Ingest chaotic telecom bills, resolve pricing entities, and generate negotiation leverage.

Phase 1: Ingestion & Privacy Isolation

Action: User uploads PDF of Comcast/Spectrum bill.

Logic: OCR extracts raw text. The agent instantly redacts PII (Name, Address, Account #) before passing data to the reasoning engine.

EFF/Jonas Principle: Privacy by Design. The LLM never sees the human identity, only the operational data.

Phase 2: Entity Resolution (The Senzing Layer)

Action: Mapping fragmented billing codes to real-world services.

Logic: The agent resolves entities. It knows that XF_BLAST_PRO, Performance Plus, and Internet 400 are all the exact same entity: a 400Mbps connection.

Operator Context: Just like matching UN and USAID aliases for the same refugee camp in East Africa. If you skip this, the AI hallucinates.

Phase 3: RAG Retrieval (Knowledge Graph)

Action: Query the SmartViewer Knowledge Graph.

Logic: Agent searches: "What is the current new-customer promotional rate for a 400Mbps connection in Zip Code 37203?"

Phase 4: ReAct Reasoning (Calculate & Decide)

Action: Compare resolved bill vs. RAG retrieved data.

Logic:
- Thought: The user is paying $105/mo for 400Mbps.
- Action: Check competitor pricing.
- Observation: AT&T Fiber offers 500Mbps for $65/mo in their area.
- Thought: User is overpaying by $40/mo ($480/yr).

Phase 5: Output (The Tactical Anchor)

Action: Generate the deliverable.

Logic: Agent outputs a highly specific, 3-bullet negotiation script for the user to read to Comcast retention agents.