Knowledge BOT using Google Gemini For An Oil & Gas Major
About the Customer
A Major Oil and Gas company intended to build a Minimum Viable Product (MVP) to extract information from the Well Completion Reports (WCR) and answer user questions focusing primarily on text content along with high level image-based questions.
Challenges
- Customer had a huge database of Well Completion Reports (WCR) and was finding it difficult to read the data from the reports and make sense of the information, and gather valuable insights.
- The report contained data in various formats, including tables, images (diagrams), and scanned documents (digital text, handwritten and type-writer text).
Our Solution
STL Digital built the PoC for:
- Data extraction from tables as label-value pairs.
- Create Vector embeddings for extracted data.
- RAG solution (with Gemini Models - Gemini-1.5, Gemini-1.0, Gemini Vision) of similarity match-based retrieval of data from the vector store and providing context to LLM for generating answers to questions.
The Outcomes
0
%
Productivity improvement through a turnaround time reduction
0
%
Accuracy while handling tabular content
0
%
