Multi-agent AI research · Powered by LangGraph

Research that thinks,
not just retrieves

Synth orchestrates four specialized AI agents — Search, Reader, Writer, and Critic — to deliver structured, cited research reports in minutes.

🔍Search Agent→📖Reader Agent→✍️Writer Chain→🎯Critic Chain
4
Specialized agents
< 60s
Avg. research time
5+
Sources per report
10/10
Quality scoring
Capabilities
Everything a research analyst does,
automated
Each agent is purpose-built for its role in the pipeline. No hallucinations — only verified sources.
🔍
Web Search
Tavily-powered search finds the most recent and relevant sources for any topic, filtering noise automatically.
📖
Deep Reading
BeautifulSoup scrapes full page content — not just snippets — giving the writer real context to work with.
✍️
Structured Writing
Groq-powered writer chain produces introduction, key findings, analysis, conclusion, and cited sources.
🎯
Critic Review
A separate critic chain scores the report 1–10 and gives actionable improvement suggestions.
⚡
LangGraph Orchestration
Agents run as a reactive graph — each step's output feeds the next with zero manual wiring.
📎
Source Tracking
Every URL is captured and displayed with title, favicon, and snippet for full transparency.
Process
How Synth works
01
step_01
You enter a topic
Type any research question — a market, technology, concept, or competitive landscape. Synth handles everything from here.
02
step_02
Search Agent queries the web
The Search Agent calls Tavily to find up to 5 fresh, reliable sources. It returns structured results: titles, URLs, and content snippets.
03
step_03
Reader Agent extracts content
Synth scrapes the top 5 relevant URLs for full-page content — headings, paragraphs, data — cleaned of navigation and ads.
04
step_04
Writer Chain composes the report
The Writer Chain synthesizes search results and scraped content into a structured report with introduction, key findings, and conclusion.
05
step_05
Critic Chain scores the output
A separate Critic Chain reviews the report for accuracy, depth, and structure — giving a score and specific improvement notes.