NexusFlow

It is a one-stop, real-time intelligence platform that ingests raw text from live data sources (news feeds, incident logs) and produces structured decision signals using a multi-agent AI pipeline.

Comment

Inspiration

In a world of information overload, the bottleneck is no longer access to data. It’s the latency of insight. We saw traders, SREs, and procurement officers drowning in "noise" (news, logs, reports). We wanted to build a system that doesn't just "classify" text, but converts it into a reliable signal fast enough to influence a multi-million dollar decision before the window of opportunity closes.

What it does

NexusFlow is a multi-agent intelligence platform that offers four specialized pipelines:

  • Market Intelligence: Converts news into BUY/SELL/HOLD signals with confidence scores, giving full description of the agent's thinking process.
  • Reliability Intel (Incident Analyst): Triage raw system logs into severity briefs: possible causes with confidence score, next actions and detailed thinking process of the agent's conclusion for a human to verify and take action.
  • Supply Chain Risk: Maps global news (e.g., strikes, weather) to specific port "Disruption Indices", alerting high risk routes.
  • Proactive Alerts: A scheduled "Sector Sweep" that pushes market intelligence via email twice daily, ensuring users never have to "search" for a crisis or a boom in the market.

How we built it

NexusFlow has four intelligent pipelines, each targeting a high-stakes real-world domain:

1. Market Intelligence Agent

Problem: Traders and analysts read hundreds of news articles daily to form a market view. Most signals are buried in noise.

Solution: A conversational AI agent that accepts natural language questions about any stock or sector. It fetches live news, analyzes sentiment like a Wall Street analyst, pulls real-time price data, and returns a structured BUY/SELL/HOLD signal with confidence score, risk level, and plain-English reasoning — all in one chat interaction.

2. Incident Analyst Agent

Problem: Large engineering teams deal with hundreds of incident reports simultaneously. Reading all of them wastes critical response time during outages.

Solution: The agent reads the raw timestamped logs, identifies affected systems, identifies patterns, computes a Severity Index based on keyword density, and historical frequency — backed by a RAG knowledge base of real system incidents — and auto-drafts an engineering brief with ranked root causes and next actions.

3. Procurement & Port Risk Agent

Problem: Supply chain managers cannot monitor global port disruptions, trade news, and geopolitical events fast enough to act before they impact procurement.

Solution: The agent monitors live news for port disruptions, trade policy changes, and logistics events. It maps events to affected supply routes and generates a Risk Score (0–1) with recommended procurement actions and automated email alerts with recommended actions with human intervention.

4. Market Analysis & Automated Alert System

Problem: Investors and analysts need a daily sector-wide performance briefing but cannot manually scan every stock and headline across multiple sectors every morning and evening.

Solution: NexusFlow automatically runs a twice-daily sector sweep at market open and market close. For each major sector it fetches the top headlines, classifies stocks as top-performing or low-performing, and generates a structured sector summary with BUY/WATCH/AVOID recommendations. These digests are pushed as automated email alerts to subscribed users.

  • Developed a weighted scoring algorithm: Signal = (LLM_Severity * Hub_Importance) + (Historical_Context).

Challenges we ran into

  • Processing live feeds hit API quotas quickly.
  • When news reported a country-level disruption that didn't match our port-specific data.
  • GitHub Configurations
  • Framing the correct prompt to get relevant response from LLM
  • How to extract the required text from the input by user to feed to LLM and to extract and modify the response which is relevant to the user's query.

Accomplishments that we're proud of

  • End-to-End Latency: Achieving a "News-to-Signal" time of under 10 seconds, not just making the agentic decisions a black-box but providing full transparency to the users about how the agent reached the conclusion.
  • Proactive Intelligence: Successfully automating the "Morning Sector Sweep" so the agent works even when the user is asleep.
  • Structural Reliability: Building an LLM RAG pipeline that consistently outputs valid JSON using Structured Output schemas, preventing system crashes during live replays.

What we learned

  • How to build an end-to-end Agentic AI (RAG pipeline).
  • How to take actions based on the results provided by AI(for eg: sending emails) which can help a user.

What's next for NexusFlow

  • Cross-Sector Correlation: Enabling agents to talk to each other—e.g., an Incident in a Tech hub automatically triggering a Risk spike in the Market Analyst agent.
  • Voice Integration: Using Gemini Live to allow users to ask "What's my portfolio risk?" while on the move.
  • Self-Correction: Implementing a feedback loop where the agent reviews its 24-hour-old predictions against actual market outcomes to "self-calibrate" its future scores.

Built With

Share this project:

Updates