
PXX SPARK.
The Overview
PXX SPARK is a fully autonomous AI agent designed to manage internal knowledge sharing. Unlike standard chatbots, SPARK features a self-optimizing "Reinforcement Learning" loop that analyzes user engagement (reactions & comments) to dynamically adjust its future content strategy without human intervention.
The Architecture
Instead of a single prompt, I architected a chain of specialized AI agents to ensure reliability and autonomy:
- Generator AgentDrafts 3 unique tips based on the selected topic using Google Gemini 2.5 Flash.
- Validator Agent (L2)Acts as a quality gatekeeper, reviewing drafts against strict guardrails (tone, length, policy) and selecting the single best version to publish.
- Learner Agent (RLHF)Analyzes post-performance metrics (upvotes/downvotes) to update the database state, prioritizing successful topics for future generation.
Infrastructure & Impact
To ensure data privacy and security, I containerized the entire application using Docker and deployed it on a self-hosted on-premise server. I also designed a custom PostgreSQL schema to track historical performance and prevent content repetition. Since deployment, the agent has maintained 100% uptime and successfully automated the weekly knowledge-sharing workflow.
Watch the autonomous agent in action
Credits & Acknowledgements
This project was developed during my internship at Parallaxx. The Proof of Concept and the video demonstration featured above are the exclusive property of Parallaxx. Special thanks to the Digital Team for their collaboration and guidance.
Tech Stack
- n8n (Workflow Automation)
- Google Gemini 2.5 Flash
- PostgreSQL
- Docker
- Microsoft Teams API
Key Features
- Self-Optimizing RLHF Loop
- Multi-Agent System Design
- 100% Uptime Architecture
- Automated Engagement Reporting