Skip to content

TanviDeore/HackAI-2026

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

46 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

NexusFlow — Real-Time Intelligence & Decision Signal Platform

"Don't just read the noise. Understand the signal."

Python FastAPI LangGraph Gemini


The Problem

Modern organizations are drowning in unstructured text — financial news, system logs, incident reports, procurement data, market feeds. The challenge is not collecting this data. The challenge is converting it into a reliable, actionable signal fast enough to matter.

  • A portfolio manager needs to know if a Fed announcement moves their position before the market opens.
  • An SRE needs to know if five login errors in three minutes mean a system-wide outage before customers start calling.
  • A procurement officer needs to know if port congestion in Shanghai affects their supply chain today, not next week.
  • An investor needs a sector-wide morning briefing without reading 200 articles.

NexusFlow solves this. It is a one-stop, real-time intelligence platform that ingests raw text from live data sources — news feeds, incident logs, procurement reports — and produces structured, quantified decision signals using a multi-agent AI pipeline. It also proactively pushes twice-daily sector intelligence alerts so decisions never wait for a human to go looking.


Track Relevance

"Can we convert text into a reliable signal that helps make better decisions?"

Track Requirement How NexusFlow Addresses It
Raw text as input Live news headlines, incident .txt files, procurement reports
Meaningful signal output BUY/SELL/HOLD, Severity Index, Index Score, Sector Score
Real-world decision support Trade recommendations, incident triage, supply chain alerts, sector digests
Full pipeline Ingest → Classify → Extract → Decide → Visualize → Alert
Scale Multi-source, multi-domain, multi-sector simultaneously
Proactive intelligence Twice-daily automated alerts without user prompting

What NexusFlow Does

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.

Example:

"What is the outlook for NVDA this week?"SIGNAL: BUY | Confidence: 82% | Risk: MEDIUM → Live 30-day price chart, sentiment breakdown, top 5 relevant headlines


🚨 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 text, 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.

Example:

SEVERITY INDEX: HIGH → Root Cause: SSO Gateway Timeout (91% confidence) → Historical match: Dec 2024 Auth Outage → Auto-drafted engineering summary with ordered action items


🚢 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.

Example:

"What is the current risk for procurement from Shanghai?"RISK SCORE: 74/100 | Action: Consider alternative sourcing → Disruption events with confidence interval >= 70, affected trade routes, recommended timeline


📊 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.

Sectors covered: Technology · Healthcare · Energy

Each sector digest includes:

  • Top 3 performing stocks with signal and reasoning
  • Bottom 3 underperforming stocks with risk flags
  • Sector-wide sentiment score
  • Key macro catalyst driving the sector
  • Recommended watchlist for the next session

Example Alert (9 AM digest):

NexusFlow Morning Sector Digest — March 8, 2026

📈 TECHNOLOGY — BULLISH (Score: 0.74)
  Top Performers:    NVDA ↑ BUY | MSFT ↑ BUY | META ↑ HOLD
  Underperformers:   AAPL ↓ SELL | INTC ↓ AVOID
  Key Catalyst:      Fed signals rate pause, AI infrastructure spending up
  Watchlist:         NVDA, MSFT

🏥 HEALTHCARE — NEUTRAL (Score: 0.12)
  Top Performers:    LLY ↑ BUY | JNJ ↑ HOLD
  Underperformers:   PFE ↓ SELL
  Key Catalyst:      FDA approval pipeline mixed signals
  Watchlist:         LLY

⚡ ENERGY — BEARISH (Score: -0.41)
  Top Performers:    None
  Underperformers:   XOM ↓ SELL | CVX ↓ AVOID
  Key Catalyst:      Oil inventory build, demand concerns
  Watchlist:         Monitor for reversal

System Architecture

┌──────────────────────────────────────────────────────────────────────┐
│                          React Frontend                              │
│   Market Intel | Incident Analyst | Port Risk | Sector Dashboard     │
└────────────────────────────────┬─────────────────────────────────────┘
                                 │ REST API (HTTP/JSON)
                                 ▼
┌──────────────────────────────────────────────────────────────────────┐
│                         FastAPI Backend                              │
│  /market/chat  /incident/analyze  /port/risk  /market-analysis/digest         │
└──────┬──────────────┬──────────────────┬──────────────┬─────────────┘
       │              │                  │              │
       ▼              ▼                  ▼              ▼
┌──────────┐  ┌──────────────┐  ┌─────────────┐  ┌────────────────┐
│ Market   │  │  Incident    │  │  Port Risk  │  │ Market Analysis│
│ Pipeline │  │  Pipeline    │  │  Pipeline   │  │ Pipeline       │
└────┬─────┘  └──────┬───────┘  └──────┬──────┘  └───────┬────────┘
     │                │                 │                  │
     ▼                ▼                 ▼                  ▼
┌─────────┐   ┌──────────────┐   ┌──────────┐   ┌────────────────┐
│ Intent  │   │ Text         │   │ Input    │   │ Market Sweep   │
│         │   | Parser Tool  │   │ location |   │                │
└─────────┘   └──────────────┘   └──────────┘   └────────────────┘
     │                │                 │                  │
     ▼                ▼                 ▼                  ▼
┌─────────┐   ┌──────────────┐   ┌──────────┐   ┌────────────────┐
│News +   │   │ Extraction   │   │  News    │   │ Per-Sector     │
│Price    │   │              │   │ Fetcher  │   │ News Fetcher   │
│Fetcher  │   └──────────────┘   └──────────┘   └────────────────┘
└─────────┘          │                 │                  │
     │          ┌────┴──────┐          ▼                  ▼
     ▼          │ RAG Tool  │   ┌──────────┐   ┌────────────────┐
┌─────────┐     │ (Chroma)  │   │  Risk    │   │ Performance    │
│Stock Risk     └────┬──────┘   │  Score   │   │ Classifier     │
│AnalysisAgent       │          │          │   |    Agent       │
└─────────┘          ▼          └──────────┘   └────────────────┘
     │        ┌──────────────┐        │                  │
     ▼        │ Root Cause   │        ▼                  ▼
┌─────────┐   │   Analysis   │  ┌──────────┐   ┌────────────────┐
│ Recomm. │   └──────────────┘  │  Action  │   │Recommendation  │
│ Agent   │          │          │  Agent   │   │                │
└─────────┘          ▼          └──────────┘   └────────────────┘
                ┌──────────────┐                          │
                │ Action       │                          ▼
                │ Planner      │               ┌────────────────┐
                └──────────────┘               │ Alert Composer │
                      │                        │                │
                      ▼                        └───────┬────────┘
                ┌──────────────┐                       │
                │ Summarizer   |                       ▼
                │              │              ┌────────────────┐
                └──────────────┘              │ Email Dispatch │
                                              └────────────────┘
                      │
                      ▼
          ┌───────────────────────┐
          │  Gemini 2.5 Flash LLM │
          │  (all agents)         │
          └───────────────────────┘

Pipeline Flows

📈 Market Intelligence

User Question (natural language)
        │
        ▼
[Intent extractor] — ticker, sector, time window
        │
   ┌────┴────┐
   ▼         ▼
[News      [Price
 Fetcher]   Fetcher]
 NewsAPI    yfinance
   └────┬────┘
        ▼
[Stock Risk Analysis Agent]
WHO/WHAT/WHY/HOW analysis per headline
        │
        ▼
[Recommendation]
BUY / SELL / HOLD
Confidence + Risk level + Reasoning
        │
        ▼
React UI — chat reply + price chart + signal card

🚨 Incident Analyst

Live Log Data (I/P from User)
        │
        ▼
reads raw text
        │
        ▼
[Extraction]
Affected systems, symptoms, timeline, severity
        │
        │
        ▼                   
    [RAG Tool]        
    Chroma vector       
search top 3 similar incidents with confidence score >= 70%
        │
        ▼
Ranked causes + confidence %
Grounded by RAG historical context
        │
        ▼
Ordered next steps
        │
        ▼
 Engineering brief
        │
        ▼
React UI — severity meter + root causes + summary

🚢 Port Risk

User Query (port / region / commodity)
        │
        ▼
[Input] — port, region, trade route
        │
        ▼
[News Fetcher] — live port/trade/geopolitical news
        │
        ▼
[Risk Scoring Agent] — Signal = (LLM_Severity * Hub_Importance) + (Historical_Context)
        │
        ▼
[Action] — switch supplier / delay / hedge
        │
        ▼
React UI — risk score + affected routes + recommendations

📊 Sector Alert (Automated, Twice Daily)

Scheduler — 9 AM + 5 PM trigger
        │
        ▼
[Sector Sweep]
Loops through 3 sectors
        │ 
        ▼
[News Fetcher Tool]
Top headlines per sector — NewsAPI
        │
        ▼
[Performance Classifier Agent]
TOP PERFORMING / LOW PERFORMING / NEUTRAL
per stock based on headline analysis
        │
        ▼
[Recommendation Agent]
BUY / WATCH / AVOID per stock
Sector sentiment score + key macro catalyst
        │
        ▼
[Alert Composer]
Formats structured sector digest
        │
        ▼
[Email Dispatch]
Sent to all subscribed users

RAG Architecture

INGESTION — one-time setup
────────────────────────────────────────────────────
Sources:
  -  Historical Incidents     (incidents.txt)
  -  IOSCO Market Outage Reports (txt)
  -  AI Incident Database     (CSV — Kaggle)
  -  Past Port Disruption Events (txt)
        │
        ▼
[TextLoader] — LangChain reads all .txt sources
        │
        ▼
[RecursiveCharacterTextSplitter]
  chunk_size = 800
  chunk_overlap = 100
  separator = "---"
        │
        ▼
[Gemini text-embedding-004]
  768-dimensional dense vectors
        │
        ▼
[ChromaDB PersistentClient]
  Collection: "financial_incidents"
  Stored at: ./chroma_db


RETRIEVAL — at query time
────────────────────────────────────────────
Raw incident text (user input)
        │
        ▼
[Gemini text-embedding-004] — embeds query into 768-dim vector
        │
        ▼
[ChromaDB cosine similarity search]
Returns top 3 most similar historical incidents
        │
  ┌─────┴──────┐
  ▼            ▼
>= 0.70       < 0.70
MATCH         NEW INCIDENT
(pattern      (novel issue,
 confirmed)    estimated)
  └─────┬──────┘
        ▼
Context passed to Gemini analyzer node
→ Root causes grounded in historical precedent
→ Confidence scores reflect match quality

Property Detail
Role Autonomous sector analyst & email alert publisher
Trigger Background threading.Timer — every 12 hours
Perceives Live sector news via NewsAPI (Technology, Finance, Energy)
Reasons Gemini 2.5 Flash batch-analyzes all sector headlines
Acts Dispatches HTML email report — no user prompt needed
Signal BUY / WATCH / SELL per sector + plain-English reasoning
Fallback Logs the full report to console if SMTP is not configured
Manual override POST /api/market/alerts/send-email from the frontend

Agent Summary

# Agent Trigger Signal Decision Gate
1 Market Intelligence User chat message BUY / SELL / HOLD Pydantic-validated Gemini enum
2 Incident Analyst User submits log text Severity + root causes (confidence %) RAG cosine similarity ≥ 0.70
3 Supply Chain Domino User queries a country Disruption Index 0.0–1.0 index > 0.75 → email modal fires
4 Sector Intelligence Autonomous every 12 hours BUY / WATCH / SELL per sector Runs on timer — no human required

All agents powered by Gemini 2.5 Flash · Agents 1–3 orchestrated by LangGraph · Agent 4 runs on autonomous background scheduler

image
image (1) image (2) image (3) Screenshot 2026-03-08 at 07 12 06

Demo Video: https://youtu.be/OV4_fNzh594

About

My contribution to the HackAI hackathon project: Developed an AI system that analyzes logs and queries large KEDB (Known Error Database) documents to retrieve relevant past incidents. If an exact match is not found, the system shows related knowledge and suggested troubleshooting steps to help engineers resolve issues faster.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • Python 50.2%
  • JavaScript 48.3%
  • CSS 1.3%
  • HTML 0.2%