NetSentient

Inspiration

Network congestion often throttles critical tasks because standard routers treat all data packets equally. We wanted to build a highly adaptive system that treats bandwidth as a precious, smart resource—allocating speed based on the actual semantic urgency and context of a user's input rather than simple, dumb data queues.

What it does

NetSentient is an intelligent bandwidth allocation platform that dynamically prioritizes data transfer speeds using a multi-tiered evaluation pipeline: 1) Tiered Routing Pipeline: Incoming inputs instantly check a rapid Cache for immediate bandwidth allocation. If missed, it processes through custom Keyword Checkers (handling complex negative rules). 2) Gemini AI Criticality Engine: Ambiguous inputs are analyzed by Gemini AI to receive a nuanced "criticality score." If the AI returns anomalies, the system automatically sanitizes and normalizes the scores based on auto-detected categories and pre-defined range safety values. 3) Fail-Safe Redundancy: If the AI encounters network issues, invalid keys, or token exhaustion, a robust Fallback Engine kicks in, routing the input through an expanded secondary keyword array to guarantee successful, uninterrupted data transfer. 4) Control & Analytics Dashboard: Users can toggle semantic routing on/off at will. The dashboard fetches low-level system metadata directly from the OS to trace live latency, bandwidth allocation percentages, and Mbps speeds alongside clean, visual graphs tracking whether inputs went via Cache, Keywords, or AI.

How we built it

1) Backend Architecture: Built with Flask and Python to manage the routing logic, fallback mechanisms, API integrations, and system-level data retrieval. 2) Frontend Experience: Designed a clean, highly interactive dashboard using HTML5, Tailwind CSS, and JavaScript to prevent data overwhelm. 3) System Integration: Leveraged psutil to hook directly into the operating system for real-time metadata collection.

Challenges we ran into

1) API Reliability & Edge Cases: AI APIs can fail due to token limits or network drops. We had to architect a bulletproof secondary keyword pipeline so data transfer never stalls. 2) Halting Bad AI Outputs: Gemini occasionally assigned erratic criticality scores. We overcame this by building an automated classification layer that enforces strict range-bound restrictions depending on the input's context. 3) System Metrics Isolation: Extracting deep semantic meaning straight from OS processes is virtually impossible. We had to carefully separate raw OS metadata from our custom semantic traffic data to keep dashboard tracking perfectly accurate.

Accomplishments that we're proud of

1) Zero-Downtime Fallback Architecture: Creating a network router that seamlessly downgrades from advanced AI scoring to macro-keywords without dropping user packets. 2) Aggregated Data Visualization: Condensing complex variables (latency, Mbps, bandwidth percentages, routing sources) into a clean, non-overwhelming UI.

What we learned

1) Defensive AI Engineering: You cannot trust raw LLM outputs blindly for real-time utility infrastructure; you must wrap AI logic in strict deterministic boundaries and fallback guardrails. 2) Resource Optimization: Balancing fast local operations (cache/keywords) with heavier remote operations (Gemini) is essential for keeping routing latency to an absolute minimum.

What's next for NetSentient

1) Local Light-LLM Deployment: Integrating a local model to replace the keyword fallback tier when offline, keeping semantic capabilities intact without internet. 2) Predictive Bandwidth Scaling: Using historical input logs to dynamically adjust the keyword databases and range caps automatically based on user behavior patterns. Conversion to Browser Extension/Application to get semantic data

Built With

• flask
• gemini
• html
• javascript
• psutil
• python
• tailwindcss