The Problem
Managing unstructured consultation notes and inbound client leads often takes hours of manual review. Relying on public cloud AI APIs to parse sensitive notes introduces data privacy risks, potential data leakage, and recurring API costs.
The Solution
I built a private, local-first triage agent that automatically ingests raw meeting notes or audio transcripts, parses key metadata (client name, core requirements, budget, priority level), and outputs formatted JSON ready for immediate logging—all while keeping sensitive data secure.
Architecture & Schemas
- Ingestion: Raw text prompt or consultation transcript.
- Processing Engine: Relevance AI node pipeline configured for schema-based extraction.
- Output: Structured JSON containing prioritized action items and client details.
Demonstration & Proof of Work
Watch the live demonstration of the triage workflow in action:
Why On-Device AI Workflows Matter
Building local-first AI pipelines ensures complete ownership over data privacy, protecting confidential client information while eliminating ongoing cloud infrastructure fees.
Top comments (2)
On-device is a sensible call for client notes. A complementary idea: much of the "unstructured" problem can be reduced at intake. If the enquiry form asks need, timeline and budget as separate questions and follows up on vague answers, the triage agent starts from cleaner input. That's what I work on at chatform.in. Would your agent accept structured answers through a webhook alongside the free-text notes?
Thanks for the feedback! Yes, the agent is configured to accept structured JSON payloads via webhook for fields like need, timeline, and budget, while keeping the free-text notes attached as supplementary context.