docs: rewrite README for ARIA

README.md

@@ -1,54 +1,66 @@
-# Glitch Squad Biomedical Assistant
+# ARIA — Autonomous Research Intelligence Agent
 
-A biomedical AI research assistant that retrieves live literature from PubMed and synthesises responses using a LangGraph ReAct agent powered by Llama 3.2.
-
-Built for the AMD Developer Hackathon 2026.
+> Multi-agent biomedical literature synthesis · AMD Developer Hackathon 2026
+> **Team Glitch Squad** — Ziauddin University, Karachi, Pakistan
 
 ---
 
 ## What It Does
 
-You type a biomedical research question. The agent autonomously searches PubMed for relevant abstracts, reasons over the retrieved literature, and returns a synthesised answer with PMID citations.
+You type a clinical research question. ARIA runs a 4-stage multi-agent pipeline that searches PubMed, synthesises evidence, scores confidence, and returns a structured report — all in under 60 seconds.
 
----
+## Pipeline
 
-## Tech Stack
+```
+Query Architect  →  Literature Scout  →  Evidence Synthesiser  →  Citation Builder
+(Groq LLaMA-3.1)    (PubMed Entrez)      (Groq LLaMA-3.1)        (structured refs)
+```
 
-| Component | Technology |
-|---|---|
-| LLM | Llama 3.2 via Ollama (local) / Mistral 7B on AMD MI300X (cloud) |
-| Agent Framework | LangGraph prebuilt ReAct agent |
-| Literature Retrieval | BioPython Entrez / NCBIPubMed API |
-| Web Framework | Flask |
-| Frontend | HTML, CSS, JavaScript |
-| Runtime | Python 3.12, WSL2 Ubuntu 24.04 |
+1. **Query Architect** — generates 5 MeSH-optimised PubMed queries from your clinical question
+2. **Literature Scout** — fetches and deduplicates papers across all queries in parallel
+3. **Evidence Synthesiser** — builds a structured 6-section synthesis with inline PMID citations
+4. **Citation Builder** — formats references with authors, journal, year, and PMID
 
----
+## Features
 
-## How To Run Locally
+- Real-time SSE streaming — progress bar updates as each stage completes
+- Structured synthesis — Background, Key Findings, Level of Evidence, Conflicting Evidence, Research Gaps, Clinical Implications
+- Evidence confidence scoring — green/yellow/red badges rated by a second LLM pass
+- Abstract viewer — click any citation to expand the full abstract inline
+- PDF export — download a formatted report with synthesis and references
 
-### Prerequisites
-- Python 3.12
-- Ollama installed
-- llama3.2 model pulled: `ollama pull llama3.2`
+## Tech Stack
 
-### Setup
+| Component | Technology |
+|-----------|------------|
+| LLM | Groq LLaMA-3.1-8B-Instant |
+| Agent Framework | LangGraph + LangChain |
+| Literature Retrieval | BioPython Entrez / PubMed NCBI |
+| Web Framework | Flask with SSE streaming |
+| PDF Generation | ReportLab |
+| Frontend | HTML, CSS, JavaScript |
+| Runtime | Python 3.12, WSL2 Ubuntu 24.04 |
+| Roadmap | AMD MI300X + Mistral 7B |
 
-    git clone https://github.com/azlaan428/glitch-squad-biomedical-assistant.git
-    cd glitch-squad-biomedical-assistant
-    python3 -m venv _venv
-    source venv/bin/activate
-    pip install -r requirements.txt
+## Setup
 
-### Run
+```bash
+git clone https://github.com/azlaan428/glitch-squad-biomedical-assistant.git
+cd glitch-squad-biomedical-assistant
+python3 -m venv venv
+source venv/bin/activate
+pip install -r requirements.txt
+export GROQ_API_KEY=your_key_here
+```
 
-    ollama serve &
-    python app.py
+## Run
 
-Open your browser at http://localhost:5000
+```bash
+./run.sh
+```
 
----
+Open **http://localhost:5000**
 
-## Team
+## Disclaimer
 
-**Glitch Squad** -- Ziauddin University, Karachi, Pakistan
+AI-generated synthesis — always verify against primary sources before clinical use.