DeepSearch Auto
An autonomous AI research platform that replicates the capability of enterprise deep-research tools like Perplexity AI. Built to showcase production-grade AI engineering, full-stack architecture, and autonomous agent systems.
Project Overview
DeepSearch Auto is a portfolio-grade autonomous research platform built to demonstrate mastery of modern AI engineering. It combines a FastAPI backend with LangGraph-orchestrated AI agents that autonomously plan and execute multi-step research workflows.
The system accepts complex research questions, breaks them into targeted sub-questions, performs parallel web searches, scrapes and processes source content, and uses Claude AI to synthesize a comprehensive, citation-backed report.
Real-time progress streaming via WebSockets keeps users informed at every step, while the modular architecture ensures the platform is scalable, maintainable, and ready for production deployment.
Key Features Added
Interactive Citation Graph
A visual node map illustrating exactly how sources connect to sub-topics and claims using React Flow.
Knowledge Graph Visualization
Dynamically maps extracted entities and their relationships (e.g., organizations, concepts) in an interactive UI.
Topic Intensity Heatmap
Clustered visualization of the most frequently mentioned keywords across all processed sources.
Confidence Dashboard
Real-time metrics tracking source agreement, evidence strength, and hallucination risk.
Professional Exports
Instantly converts complex research outputs into downloadable PDF, DOCX, and Markdown formats.
Session History
Safely stores your active research session data locally for quick review, automatically clearing upon logout.
Research Pipeline
How the autonomous agent processes your query
Research Planning
Claude AI analyzes the query and generates 5 targeted sub-questions covering different angles of the topic.
Parallel Search
SerpAPI executes simultaneous Google searches for each sub-question, collecting top-ranked URLs and metadata.
Content Extraction
Trafilatura scrapes each URL concurrently, extracting clean readable text while removing ads and navigation.
AI Synthesis
Claude synthesizes all gathered evidence into a structured report with executive summary, findings, and conclusion.
Report Generation
The final report is streamed to the frontend with confidence scores, source citations, and export options.
Technology Stack
Enterprise-grade tools and frameworks
Frontend
Backend
AI & Search
Infrastructure
System Architecture
Clean, modular, production-ready design
Developed by
Pranav V P
AI Engineer & Full Stack Developer specializing in autonomous agent systems, LLM orchestration, and production-grade AI applications. Passionate about building tools that push the boundaries of what's possible.