🦜🔗 Track Package Delivery w. Langchain Agent Custom Tools

adriens - Aug 24 '23 - - Dev Community

☝️ About

Big enteprises build Strategic Plans to project themselves into the future.

OPT-NC makes no exception and has recently designed its own called "Construire demain":

Plan stratégique | OPT-NC

L’OPT-NC a présenté l’élaboration du plan stratégique CONSTRUIRE DEMAIN | OPT 2025 à son conseil d’administration le 16 novembre 2021. Un choix d’orientation a été fait à cette occasion, soutenu par des échanges préalables au gouvernement de la Nouvelle-Calédonie et par de multiples contributions de personnalités politiques, économiques et sociales dans le pays.  

favicon office.opt.nc

A strategic plan may have a lot (maybe plenty) of points... and in the mean time should be resumable into a very few keypoints -let's say 3 -,

so anyone can understand and talk about the big picture with ease.

👉 With this post, we'll dive into a journey where, with the help of various AI driven tools, we will:

  1. 📚 Chat with the plan itself
  2. 🤔 Imagine a near future
  3. 💡 Pitch an idea
  4. 🗣️ Explain & Implement the idea
  5. 🕹️ Play with the prototype
  6. 🎯 Loop back on strategic plan to how it fits
  7. 🔭 Think about other opportunities

💭 About Chatbots & Primary Customer Service

Below some context about ChatBot based service, good to knonw before delivering such services in production:

Image description

🧰 Tools

We will use many tools here:

  • 🧠 Quivr as "the brain" of the Strategic Plan, which holds the knowledge of what the plan we have stored in
  • 🦾 OpenAI : the API that will give us the LLM, the engine that will "do the job"
  • 🤖 ChatGPT : for casual question or little explanations (eg. "What's a LLM ?")
  • 🦜🔗 Langchain : the framework, the "glue" we will use to interact with the LLM
  • 😽 Kor : A tool to extract structured data out raw unstructured text

🎯 What we'll do

In this demo, we'll see how to build very custom tools (add our very own ones) to achieve specific customer-centric tasks... but operated by autonomous AI-driven agent.

At the end, we'll get a kind of connected chatGPT.

📽️ Demo

🤯 What Ive learned and reflexions

  • Trust LLM capabilities (experiment early with tools)
  • Screenwriter role
  • Customer Experience Designer
  • Designing Custom Tools
  • About static content : fallback
  • Agent Contextualizaiotn
  • Agent design
  • Agent & tools unit testing
  • Chatbot KPIs

📚 Must-reads

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .