- Built w## Inspiration As a data science student, I constantly waste time manually browsing 10+ websites just to compare developer tools. I wanted an AI agent that could do this research automatically and return structured, reliable results — like having a senior engineer do the research for you.
What it does
ERA (Engineering Research Assistant) is an AI agent that takes a developer query like "best Firebase alternatives" and automatically:
- Searches and scrapes the web using Firecrawl
- Extracts the top relevant tools from articles
- Researches each tool's official website
- Returns structured cards with pricing, tech stack, language support, API availability
- Generates an expert recommendation at the end
How we built it
- LangGraph — stateful 3-node workflow: extract tools → research each tool → analyze and recommend
- Firecrawl — web search and page scraping, returns clean markdown
- OpenAI gpt-4o-mini — LLM with Pydantic structured outputs to prevent hallucination
- Pydantic — enforces structured data schema from every LLM response
- Streamlit — minimal developer-focused UI inspired by Linear and Vercel Dashboard
Challenges we ran into
- Getting the LLM to return consistent structured output across wildly different tools
- Preventing hallucination by grounding every response in actual scraped content
- Designing a UI that feels like a developer tool, not an AI startup landing page
- Handling cases where Firecrawl returns incomplete or irrelevant content
Accomplishments that we're proud of
- Built a fully stateful multi-step agent from scratch using LangGraph
- Achieved structured, reliable output using Pydantic even for messy web content
- Designed a clean minimal UI that genuinely looks like a professional developer tool
- The agent correctly identifies pricing models, open source status, and integrations purely from scraped homepage content
What we learned
- How stateful LangGraph workflows differ from simple chatbot loops
- How Pydantic enforces structured LLM outputs and prevents hallucination
- How Firecrawl returns clean LLM-ready markdown from any webpage
- How to design developer tools that prioritize information density over aesthetics
What's next for ERA — Developer Tools Research Agent
- GitHub repository analyzer — paste a repo URL and get tool recommendations based on your actual codebase
- Stack compatibility checker — flags when recommended tools don't integrate well together
- Side-by-side comparison table — compare all researched tools in one view
- Markdown export — download the full research report
- Search history — save and revisit previous research sessions
Built With
- firecrawl
- langchain
- langgraph
- openai
- pydantic
- python
- streamlit
