Multi-agent research engine orchestrating 5 specialized LangGraph agents. Decomposes market queries, executes parallel web scraping with Tavily, verifies claims with local LLMs, and synthesizes publication-ready PDF reports.
Comprehensive market research is fragmented and labor-intensive. Analysts spend hours manually browsing dozens of industry reports, cross-checking conflicting statistics, weeding out sponsored clickbait, and formatting summaries into presentation decks. Single-prompt AI tools often hallucinate unsourced figures or fail to maintain citation traceability across broad research queries.
An autonomous multi-agent workflow built on LangGraph. Five specialized agents collaborate deterministically: a Planner decomposes queries into sub-topics, Search and Scraper agents gather live web data via Tavily and BeautifulSoup, a Verifier validates quantitative claims using local Ollama (qwen2.5:7b), and a Summarizer synthesizes executive insights via Groq (llama-3.3-70b)—complete with interactive ChromaDB RAG and automated ReportLab PDF exports.
Planner, Search, Scraper, Verifier, and Summarizer agents coordinate seamlessly through a stateful LangGraph execution engine.
Combines local privacy & cost efficiency (Ollama qwen2.5:7b for planning & verification) with high-speed cloud reasoning (Groq llama-3.3-70b for synthesis).
Leverages Tavily search API and httpx + BeautifulSoup web scrapers to parse live source articles, news, and market data.
Embeds scraped research articles into ChromaDB, allowing users to ask follow-up questions directly grounded in fetched source context.
Generates publication-ready, multi-page executive PDFs containing executive summaries, key trends, source citations, and visual charts.
Includes pre-configured n8n workflows for Telegram trigger bots, automated daily re-research, and email PDF delivery.
agents/planner.py — Decomposes complex queries into focused research sub-tasksagents/search.py — Executes Tavily search queries for targeted market domainsagents/scraper.py — Resilient HTML content extractor stripping site clutteragents/verifier.py — Evaluates source credibility and cross-verifies statistical claimsagents/summarizer.py — Synthesizes structured JSON reports using Groq 70Bgraph/workflow.py — LangGraph StateGraph connecting agent state transitionsdb/chroma_client.py — ChromaDB vector store for semantic interactive Q&Areport/pdf_builder.py — ReportLab engine generating styled PDF reportsn8n/ — Automation JSONs for Telegram triggers and automated email deliveryStateful multi-agent orchestration framework for complex cyclic AI graph workflows.
Hybrid local LLM (qwen2.5:7b) and ultra-fast cloud LLM inference API (llama-3.3-70b).
Production REST API backend with Swagger docs and an intuitive 4-tab Streamlit dashboard.
Vector store for conversational document RAG and relational database for research history.