ARIA - Autonomous Research Intelligence Assistant

An intelligent two-phase research system: from classical NLP to a fully autonomous AI research agent.

---

Overview

ARIA is built as a two-milestone project that demonstrates the progression from traditional machine learning and NLP to modern agentic AI workflows.

Milestone	Approach	Status
M1 — NLP Analysis	Classical ML: TF-IDF, LDA, TextRank	Complete
M2 — ARIA Agent	LangGraph + Groq (Llama 3.3 70B) + DuckDuckGo	Complete

---

Quick Start

1. Clone & Setup (one command)

git clone https://github.com/oksaumya/ARIA.git
cd ARIA
bash setup.sh

setup.sh creates an isolated .venv virtual environment, installs all dependencies, and downloads the required spaCy and NLTK models automatically.

2. Add your Groq API Key

Open .streamlit/secrets.toml and add your key:

GROQ_API_KEY = "gsk_your_key_here"

Get a free key at console.groq.com — no credit card required.

3. Run the App

source .venv/bin/activate
streamlit run app.py

Open http://localhost:8501 in your browser.

---

Manual Setup (alternative to setup.sh)

# Create and activate virtual environment
python3 -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Download language models
python -m spacy download en_core_web_sm
python -c "import nltk; nltk.download('punkt'); nltk.download('punkt_tab'); nltk.download('stopwords')"

# Run
streamlit run app.py

---

Project Structure

ARIA/
  app.py                          # Home page (Streamlit multi-page entry)
  setup.sh                        # One-shot environment setup script
  requirements.txt                # All Python dependencies
  pages/
    1_NLP_Analysis.py             # Milestone 1 — NLP pipeline UI
    2_AI_Research_Agent.py        # Milestone 2 — ARIA agent UI
  src/
    data_fetcher.py               # PDF / DOCX / TXT / Wikipedia extraction
    nlp_pipeline.py               # TF-IDF, LDA, TextRank
    agent/
      state.py                    # LangGraph ResearchState TypedDict
      graph.py                    # Graph topology + conditional retry edges
      nodes.py                    # 7 node functions (plan → search → fetch → summarize → aggregate → validate → report)
      llm.py                      # Groq API client (Llama 3.3 70B)
      web_search.py               # DuckDuckGo wrapper with retry
      report.py                   # Markdown renderer + fpdf2 PDF export
      memory.py                   # Session history (st.session_state)
      qa_agent.py                 # Follow-up Q&A grounded in report
  .streamlit/
    secrets.toml                  # API keys (gitignored — never commit)

---

Milestone 1 - NLP Analysis System

Objective: Analyze research documents using classical NLP without any LLMs.

Features:

• Upload PDF, DOCX, or TXT files — or search Wikipedia directly
• Extractive Summary via TextRank (Sumy)
• Keyword Extraction via TF-IDF with bar chart visualization
• Topic Modeling via LDA (4 topics, configurable)
• Raw text preview

Tech: spaCy · scikit-learn · Sumy · Gensim · PyMuPDF

---

Milestone 2 - ARIA Agent

Objective: Autonomously research any topic by searching the web, synthesizing multiple sources, and generating a structured report.

LangGraph Workflow:

plan_research → web_searcher → content_fetcher → source_summarizer
             → aggregator → quality_validator
                                   |
                    +--------------+---------------+
                    |                              |
              score < 0.6                   score >= 0.6
              retry_count < max             (or max reached)
                    |                              |
             web_searcher (retry)          report_generator → END

Features:

• Query planning :- Grok decomposes query into 3–5 targeted sub-questions
• Web search :- DuckDuckGo (free, no API key needed for search)
• Content fetching :- Fetches and parses full page text via BeautifulSoup
• Per-source AI summaries :- Groq summarizes each source in context
• Iterative quality control :- Re-searches if coverage score < 60%
• Structured report :- Title · Abstract · Key Findings · Conclusion · Sources
• Follow-up Q&A :- Chat interface grounded in the report (no hallucination beyond sources)
• PDF export :- Full report + Q&A history
• Session history :- Last 10 research sessions in sidebar

Tech: LangGraph · LangChain · Groq (Llama 3.3 70B) · DuckDuckGo Search · fpdf2

---

Deployment (Streamlit Community Cloud)

1. Push to GitHub
2. Go to share.streamlit.io → New app → select app.py
3. Under Advanced settings → Secrets, add:

   GROQ_API_KEY = "gsk_your_key_here"

4. Deploy — Streamlit Cloud installs requirements.txt automatically

---

Constraints & Requirements

• Team Size: 3 Students
• API Budget: Free Tier Only (Groq free tier · DuckDuckGo no API key)
• Framework: LangGraph (mandatory)
• Hosting: Streamlit Community Cloud

---

Technology Stack

Component	Technology
LLM	Groq API — Llama 3.3 70B Versatile (free tier)
Agentic Framework	LangGraph
Web Search	DuckDuckGo Search (no API key)
NLP (M1)	spaCy · scikit-learn · Sumy · Gensim · NLTK
UI	Streamlit (multi-page)
PDF Export	fpdf2
Document Parsing	PyMuPDF · python-docx

---

Evaluation

Phase	Weight	Criteria
Mid-Sem (M1)	25%	NLP Pipeline correctness, Topic Modeling quality
End-Sem (M2)	75%	Agentic workflow, report quality, UI, deployment