Case study

EnviroGPT

A chatbot for NEOM's environment teams that answers questions from the organization's environmental documentation, using retrieval-augmented generation.

For: NEOM
My role: Architect & builder
Stack: GenAI, RAG, Chatbot, Palantir Foundry

Problem

NEOM's environment teams relied on a large body of environmental documentation, and finding the right answer meant searching through files by hand.

What I built

A RAG chatbot on Palantir Foundry: the relevant environmental documents are indexed, and each answer is generated from the passages retrieved for the question.

Outcome

Environment teams get answers grounded in NEOM's own documentation directly in chat, instead of digging through files.

Architecture

01Environmental docsRelevant documentation collected
02IndexingDocuments indexed for search
03RetrievalMost relevant passages per question
04LLM answerResponse generated from those passages
05ChatUsed by NEOM environment teams

Grounded in NEOM documents
Built on Foundry

The problem

NEOM's environment teams work with a large body of environmental documentation. Answering a question meant knowing which file to open and then searching through it by hand.

What I built

EnviroGPT is a chatbot that answers those questions from the documentation itself:

Indexing. The relevant environmental documents are indexed so they can be searched.
Retrieval. For each question, the most relevant passages are pulled from the index.
Generation. An LLM writes the answer in chat from those passages, so responses are grounded in NEOM's own documents rather than general knowledge.

It was built on Palantir Foundry alongside NEOM's other environmental data work.

Who used it

NEOM's environment teams, as a faster way to get answers from their documentation.

Details are kept high-level because this is internal NEOM work.

Internal NEOM work, shared at a high level without confidential data.

Next case studyTender360