Building and Evaluating a Company Research Agent

[@smalyu]

At 13:40, there’s an important point to consider: with structured output, LLMs generate data strictly in the order defined by the schema. In this case, score comes before reason, which can reduce accuracy because the model is forced to assign a numerical score first and only then justify it with a reason. This approach limits the model’s ability to evaluate thoroughly. Reversing the order, so the model generates its reasoning (reason) first and then derives the score (score) based on that reasoning, would likely result in more accurate evaluations.

[@austinsand8163]

Thank you for including the example of an evaluation tool for this, it was really helpful to see and to be reminded of the importance of that cycle.

[@AlexJohnson-g4n]

Great resource for enriching data using AI. Tavily is powerful for company research. Looking forward to its evolution.

[@albertgilopez-ytt]

Thank you for sharing this amazing content! 🤓

[@anshultibrewal5226]

This is a really good video! Thanks for teaching this concept

[@olwiba]

Great video, thanks for sharing!

[@fpingham]

Fantastic Lance!!

[@pavellegkodymov4295]

Great, thanks! For evaluation, where do expected values are coming from? From a raw initial collected text?

[@dmytrobilyi14]

Please, help with this problem "TypeError: No synchronous function provided to "research_company".Either initialize with a synchronous function or invoke via the async API (ainvoke, astream, etc.)"

[@developer-h6e]

Any idea for one that helps searxh for scholarships, in specific site like .edu etc custome one so that result are only those. Then searxh in result pages looking if i meet criteria , amount, deadline, etx. That would really help.

[@RohanKumar-vx5sb]

wow moving at ligtening speed

[@kandukuriadityakumar771]

Please create an agentic ai application with langchain, it will be more helpful.