Naija Data Intelligence Assistant 🇳🇬

URL: Live App on Streamlit
Repository: GitHub

Overview

Naija Data Intelligence Assistant (9jaAssist) is a multi-agent, retrieval-augmented generation (RAG) system designed to extract, summarize, and analyze data about Nigeria from the web in response to natural language queries. Leveraging LangGraph for agent orchestration, OpenAI for language modeling, Tavily for real-time search, and Streamlit for deployment, the system provides structured answers, insight summaries, and source traceability.

Motivation & Problem Space

Despite increasing data publication from Nigerian institutions, there's a gap in accessibility, contextualization, and actionable understanding for non-technical users. Government reports are fragmented; real-time data is locked in articles or PDFs. Analysts and citizens alike struggle to:

• Query Nigerian metrics in plain English
• Retrieve up-to-date, multi-source answers
• Gain insight (not just data) from search results

Goal: Build a plug-and-play, RAG-based web assistant that answers questions like:

"What's the trend of maternal mortality in Nigeria from 2010 to 2020?"

And returns:

• Concise answer (with date validity)
• Insight bullets and trend detection
• Source citations with titles

System Architecture

User --> Streamlit UI --> LangGraph (state machine) --> Multi-Agent Flow
                                            |
                                            |--> [Tavily Search] --> Result chunks
                                            |--> [Answer Synthesis Agent] --> Final summary
                                            |--> [Insight Extraction Agent] --> Bullet pattern report

Each node in the graph is stateless but processes typed, structured memory passed across the system.

Multi-Agent Workflow (LangGraph Orchestration)

1. Search Agent (search_node)

• Uses TavilySearchResults with explicit API key

• Extracts top 3 result chunks (content, title, URL)

• Stores:

  • search_text (concatenated plain text for LLM)
  • source_links (dict with title, url)

2. Summarization Agent (summarize_node)

• LangChain PromptTemplate:

  • Asks the LLM to use search text to answer the original question concisely and factually
  • Avoids speculation or hallucination

• Uses ChatOpenAI(gpt-4) with deterministic temperature=0

3. Insight Agent (insight_node)

• Accepts the output from summarization

• Extracts:

  • Trends (e.g., "rising since 2015")
  • Patterns ("seasonal spikes in Q4")
  • Suggestions (e.g., "chart: line plot of inflation (2020-2025)")

Typed State Definition

class AgentState(TypedDict):
    question: str
    search_text: Optional[str]
    source_links: List[Dict[str, str]]  # title + url
    answer: Optional[str]
    insights: Optional[str]

This enforces schema consistency and enables traceability across LangGraph nodes.

Tech Stack Summary

Deployment Notes

config.py supports hybrid local/dev mode:

import os, streamlit as st
OPENAI_API_KEY = st.secrets.get("OPENAI_API_KEY", os.getenv("OPENAI_API_KEY"))
TAVILY_API_KEY = st.secrets.get("TAVILY_API_KEY", os.getenv("TAVILY_API_KEY"))

secrets.toml:

OPENAI_API_KEY = "sk-..."
TAVILY_API_KEY = "tvly_..."

Example Interaction

Q: What is Nigeria's maternal mortality rate?

Answer: As of 2020, the maternal mortality rate in Nigeria is 1047 deaths per 100,000 live births, one of the highest globally.

Insights:

• Nigeria has shown marginal improvement since 2015.
• WHO targets < 70 per 100,000 by 2030; Nigeria is off-target.
• Major causes include hemorrhage, sepsis, unsafe abortion.
• Suggested chart: bar comparison (Nigeria vs. Sub-Saharan average)

Sources:

• World Bank – Nigeria Mortality
• WHO Reproductive Report

Use Cases

Future Enhancements

• Support uploading CSV or PDF documents (offline + structured retrieval)
• Hybrid Retrieval: vector + web search
• Chart generation (via Plotly or Altair)
• Real-time dashboards or scheduled reports
• Admin tool for auditing sources and citations

Conclusion

NaijaAssist represents a powerful demonstration of agent-based reasoning over open-world information grounded in Nigerian data. By coupling LangGraph's modular flow with the retrieval abilities of Tavily and the summarization strength of GPT-4, this assistant bridges the gap between data and decisions for Nigerian users.

Whether you're a data scientist, student, or citizen—insight is just one question away.