Dispatch

AI agent for company research. Type a company name, get a structured brief powered by live web search and Claude.

Live: https://dispatch-agent.netlify.app Backend: https://github.com/harshgolani/dispatch/tree/main/backend

---

What it does

1. You type a company name
2. The agent runs three targeted web searches via Tavily — overview, recent news, engineering culture
3. Claude Haiku classifies each result for relevance. If irrelevant, retries with a broader query.
4. Claude Sonnet synthesizes all results into a structured markdown report
5. The execution trace shows every step — which model ran what and why

Output format:

• What They Do — products, positioning, business model
• Recent News — latest developments, funding, launches
• Engineering & Culture — stack, hiring, Glassdoor signals
• Why It's Interesting — for job seekers and investors

---

Architecture

Browser → Netlify (React) → Render (FastAPI) → Agent Pipeline

The agent pipeline runs three steps:

1. Tavily searches the web (3 targeted queries)
2. Claude Haiku classifies each result for relevance
3. Claude Sonnet synthesizes all results into a structured report

---

The routing decision

Every search result is classified by Haiku before reaching Sonnet. Classification is a binary yes/no question — cheap, fast, and Haiku handles it correctly. Sonnet runs exactly once for the final synthesis. This reduces token cost by ~60% compared to routing everything through Sonnet.

Simple tasks → Haiku. Complex reasoning → Sonnet. This is how production AI systems manage inference cost.

---

Stack

Layer	Technology
Frontend	React 19 + Vite
Backend	FastAPI + Python
Web retrieval	Tavily
Classification	Claude Haiku
Synthesis	Claude Sonnet
Frontend hosting	Netlify
Backend hosting	Render

---

Run locally

Backend:

cd backend
pip install -r requirements.txt
cp .env.example .env
# Add ANTHROPIC_API_KEY and TAVILY_API_KEY to .env
uvicorn main:app --reload
# Runs on http://localhost:8000

Frontend:

cd frontend
npm install
npm run dev
# Runs on http://localhost:5173

Update API constant in src/App.jsx to switch between local and production backend.

---

API

POST /research

Request:

{"company": "Anthropic"}

Response:

{
  "company": "Anthropic",
  "report": "## What They Do\n...",
  "trace": [
    {"step": "search_overview", "model": "tavily", "query": "..."},
    {"step": "classify_overview", "model": "haiku"},
    {"step": "classify_overview_result", "verdict": "relevant"},
    ...
    {"step": "synthesize", "model": "sonnet"}
  ]
}

---

Security

• CORS restricted to Netlify frontend URL
• Rate limiting: 10 research requests/hour per IP via slowapi
• 100 character input limit on company name
• API keys stored as environment variables, never in code

---

Known limitations

• No streaming — report appears all at once after 5-8 seconds. Streaming is a Phase 2 improvement.
• Session-only results — lost on page refresh. No database required for current use case.
• Single retry on irrelevant results — production version would use multiple query reformulation strategies.
• Render free tier spin-down — first request after inactivity takes 30-60 seconds to wake the server.