Building a Private, Local Lead & Note Triage Agent for a Freelance Colleague

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

My friend runs an independent consulting practice dealing daily with confidential client notes and unstructured inquiries. Due to strict non-disclosure agreements and data privacy requirements, sending raw client data through third-party proprietary LLM APIs poses unacceptable compliance risks.

They needed an automated system to sanitize, summarize, and prioritize inbound client notes—without the data ever leaving local hardware.

Demo

The local agent pipeline processes raw unstructured text directly on local hardware using open-weight Gemma inference, outputting validated JSON payloads:

json
{
  "metadata": {
    "timestamp": "2026-10-02T12:00:00Z",
    "source_type": "client_note",
    "privacy_level": "local_only"
  },
  "parsed_output": {
    "summary": "Client requested immediate technical audit for onboarding workflow.",
    "action_items": [
      "Review schema validation rules",
      "Schedule intake call"
    ],
    "urgency_rating": "high"
  }
}
"Having a local triage tool gives me complete confidence that sensitive notes are handled securely without risking client confidentiality." — End User Feedback

## Code

  

    

      
      
        rcortez056-spec
       / 
        hacktoberfest-2026-build-for-a-friend
      
    
    

      Privacy-first local agent workforce and schema specifications for Hacktoberfest 2026
    
  
  

    

Private Local Text Triage Agent




Built for Hacktoberfest 2026 Launch Weekend Challenge: Build for a Friend



Overview



An offline-first, privacy-focused agent pipeline designed to parse sensitive unstructured notes and client communications using open-weight Gemma inference. This setup guarantees complete data sovereignty and zero cloud model dependencies.


Key Features





Data Sovereignty: Operates strictly on local hardware with open-weight models.

Schema Validation: Strict JSON schema enforcement for downstream automation.

Task Decomposition: Separates ingestion, inference, and structured output formatting.



Repository Contents





schema.json: JSON Schema definition for inputs and outputs.

system_prompt.txt: Production prompt for local Gemma model execution.



Submission Details





Challenge: Hacktoberfest Weekend Challenge 1 (Build for a Friend)

Target Category: Best Use of Gemma

Primary Repository: rcortez056-spec







  


  
View on GitHub




  
  
  How I Built It


The system utilizes Google's Gemma open-weight model executed locally to ensure complete data sovereignty and predictable execution costs:



Ingestion & Validation: Input payloads are verified against a strict JSON Schema definition (schema.json).

Local Inference Engine: Gemma processes the raw text locally using an execution-focused prompt (system_prompt.txt).

Task Decomposition & Output Dispatch: Extracts key summaries, actionable steps, and urgency ratings (low, medium, high) formatted for local storage and downstream processing.


By leveraging open-weight models, this solution guarantees zero cloud model dependencies, zero API call fees, and strict privacy compliance.