NeuraBrief: AI-Driven Command & Crisis Intelligence Platform

• 

  Initial screenshot without running analysis

• 

  This is the main screenshot which shows the analysis

• 

  Shows the live updates in simmulation

About NeuraBrief

The Inspiration

Urban crisis response systems have traditionally relied on fragmented dashboards, requiring managers to manually sift through telemetry to make high-stakes decisions. NeuraBrief was designed to build a unified "Single Pane of Glass" terminal that extracts messy, unstructured crisis logs (via custom JSON inputs) and processes them into structured operational intelligence.

The Architecture & Engineering

We engineered NeuraBrief as a cohesive full-stack real-time intelligence hub:

• GenAI Reasoning Core: Deeply integrated with the gemini-2.5-flash model. We constructed a strict, single-pass unified JSON prompt contract. Instead of sending separate queries for different panels, the LLM performs a holistic crisis evaluation, returning coordinates for 72-hour resource forecasts $P(t)$, multi-vector risk factors $\vec{R}$, metrics, and action plans in a single API call to reduce latency: $$P(t) = P_0 \cdot \cos\left(\frac{\pi t}{24}\right) + C_{\text{surge}}(t)$$
• Frontend HUD: A Vite + React 19 application utilizing vanilla CSS to manage custom HSL glow tokens, grid overlays, and responsive breakpoints, backed by Chart.js for high-speed canvas telemetry.
• Fail-Safe Normalization Layer: Standard LLMs can occasionally return invalid schemas or trailing commentary, which could freeze client charts. We built an advanced backend parsing pipeline using jsonrepair and fallback-clamp layers on Express, ensuring that if a key is malformed, the system dynamically swaps in localized baselines for that specific component with zero downtime.

ReactJS + Vite with Express Client and Server Folder with node_modules Used render for deployment

Code Running - npm install --include=dev; npm run build

What We Learned & Overcame

Handling multi-panel coordination dynamically proved to be our biggest challenge. Directing the model to construct a radar chart where the coordinates exactly corresponded with the timeline warnings required meticulous prompt calibration. We also successfully implemented a concurrent build architecture on Render, allowing a single lightweight Node service to serve compiled client bundles alongside the secure Express API securely.

Built With

• built-with-languages:-javascript-(es6+)
• chart.js-(line-&-radar-telemetry)-backend-engine:-express.js
• css3-(custom-glassmorphic-variables)-frontend-framework:-react-19
• hsl
• html5
• jsonrepair-genai-apis:-google-gemini-api-(gemini-2.5-flash)
• neon
• node.js
• openrouter-api-design-philosophy:-cyberpunk-command-hud
• palettes
• tailored
• vite