Wanderlust, a University of Tokyo Matsuo Lab Startup, Conducts PoC on Improving RAG Accuracy Using Unstructured Data Structuring Technology in DENSO's AI Project
Key facts
- Wanderlust, a University of Tokyo Matsuo Lab Startup, Conducts PoC on Improving RAG Accuracy Using Unstructured Data Structuring Technology in DENSO's AI Project
- Wanderlust has completed a Proof of Concept (PoC) with DENSO to improve RAG accuracy by structuring unstructured data like charts using OCR and VLM, successfully addressing a major bottleneck in corporate AI adoption.
- Source: PR Times
- Date: April 6, 2026
Direct answer
Wanderlust has completed a Proof of Concept (PoC) with DENSO to improve RAG accuracy by structuring unstructured data like charts using OCR and VLM, successfully addressing a major bottleneck in corporate AI adoption.
- Citation
- Wanderlust, a University of Tokyo Matsuo Lab Startup, Conducts PoC on Improving RAG Accuracy Using Unstructured Data Structuring Technology in DENSO's AI Project (April 6, 2026), PR Times
- Source
- PR Times
- Date
- April 6, 2026
Wanderlust has completed a Proof of Concept (PoC) with DENSO to improve RAG accuracy by structuring unstructured data like charts using OCR and VLM, successfully addressing a major bottleneck in corporate AI adoption.
📋 Article Processing Timeline
- 📰 Published: April 6, 2026 at 18:00
- 🔍 Collected: April 6, 2026 at 09:01
- 🤖 AI Analyzed: April 21, 2026 at 03:50 (354h 49m after Collected)
Wanderlust Inc. (Headquarters: Chiyoda-ku, Tokyo, CEO: Hibiki Nishikawa, hereinafter "Wanderlust"), a startup originating from the Matsuo Lab at the University of Tokyo, announces that it has conducted a technical verification (Proof of Concept, PoC) specialized in processing and structuring unstructured data for the advanced internal data search infrastructure project promoted by DENSO Corporation (Headquarters: Kariya City, Aichi Prefecture, President and CEO: Shinnosuke Hayashi, hereinafter "DENSO").
In this PoC, Wanderlust utilized AI analysis technology to structure "unstructured data," which has conventionally been difficult to read in RAG (Retrieval-Augmented Generation). For complex document groups including charts and graphs, the company utilized a structuring method built with a pipeline of OCR (Optical Character Recognition) and VLM (Vision Language Model). It was confirmed that by optimizing prompt adjustments to suit DENSO's internal data, response accuracy was significantly improved.
■ Background and Purpose of the PoC
While the adoption of RAG for utilizing internal corporate knowledge is advancing in recent years, accurately interpreting chart data heavily used in marketing reports and technical documents, such as "line graphs," "bar graphs," and "scatter plots," has become a major challenge. The inability of AI to process these unstructured data was a bottleneck for response accuracy. DENSO has also been building an AI infrastructure to seamlessly utilize internal and external data, making it an urgent task to improve the reference accuracy of documents containing complex charts. To solve this issue, Wanderlust, which has strengths in unstructured data analysis, provided technical support for this project.
■ Overview of the PoC
Targeting unstructured data held by DENSO, Wanderlust conducted "structuring processing" using its technology to verify changes in response accuracy in tools like Copilot Studio.
1. Verification Details
- Structuring of Chart Data: Converting and structuring unstructured data containing specific formats (graphs, charts, etc.) into formats interpretable by AI.
- Accuracy Comparison Verification: Quantitatively measuring the difference in AI response accuracy (correctness and comprehensiveness) between the "raw unstructured data" state and the "post-structuring processing" state.
- Consideration of Optimal Architecture: Supporting the design of optimal RAG architecture for handling unstructured data.
2. Role of Wanderlust
Providing technical support and consulting in the highly technically difficult area of "unstructured data processing." By leveraging academic knowledge and implementation capabilities to improve data "quality," Wanderlust contributes to maximizing the value of the entire AI search infrastructure.
In this PoC, Wanderlust utilized AI analysis technology to structure "unstructured data," which has conventionally been difficult to read in RAG (Retrieval-Augmented Generation). For complex document groups including charts and graphs, the company utilized a structuring method built with a pipeline of OCR (Optical Character Recognition) and VLM (Vision Language Model). It was confirmed that by optimizing prompt adjustments to suit DENSO's internal data, response accuracy was significantly improved.
■ Background and Purpose of the PoC
While the adoption of RAG for utilizing internal corporate knowledge is advancing in recent years, accurately interpreting chart data heavily used in marketing reports and technical documents, such as "line graphs," "bar graphs," and "scatter plots," has become a major challenge. The inability of AI to process these unstructured data was a bottleneck for response accuracy. DENSO has also been building an AI infrastructure to seamlessly utilize internal and external data, making it an urgent task to improve the reference accuracy of documents containing complex charts. To solve this issue, Wanderlust, which has strengths in unstructured data analysis, provided technical support for this project.
■ Overview of the PoC
Targeting unstructured data held by DENSO, Wanderlust conducted "structuring processing" using its technology to verify changes in response accuracy in tools like Copilot Studio.
1. Verification Details
- Structuring of Chart Data: Converting and structuring unstructured data containing specific formats (graphs, charts, etc.) into formats interpretable by AI.
- Accuracy Comparison Verification: Quantitatively measuring the difference in AI response accuracy (correctness and comprehensiveness) between the "raw unstructured data" state and the "post-structuring processing" state.
- Consideration of Optimal Architecture: Supporting the design of optimal RAG architecture for handling unstructured data.
2. Role of Wanderlust
Providing technical support and consulting in the highly technically difficult area of "unstructured data processing." By leveraging academic knowledge and implementation capabilities to improve data "quality," Wanderlust contributes to maximizing the value of the entire AI search infrastructure.
FAQ
What company conducted the Proof of Concept for DENSO's AI project using unstructured data structuring technology?
Wanderlust Inc., a startup from the University of Tokyo Matsuo Lab, conducted the Proof of Concept for DENSO's AI project using unstructured data structuring technology.
Which university-affiliated lab is Wanderlust Inc. associated with according to the article?
Wanderlust Inc. is associated with the Matsuo Lab at the University of Tokyo, where it originated as a research-based startup.
What specific technology did Wanderlust use to process complex documents containing charts and graphs during the PoC?
Wanderlust used a pipeline combining OCR (Optical Character Recognition) and VLM (Vision Language Model) to process and structure complex documents with charts and graphs.
What improvement in RAG performance was confirmed during the collaboration between Wanderlust and DENSO?
By optimizing prompt adjustments for DENSO's internal data, Wanderlust significantly improved the response accuracy of RAG systems during the technical verification.
What types of data posed challenges for RAG systems before Wanderlust's intervention in DENSO's project?
Chart data such as line graphs, bar graphs, and scatter plots in marketing reports and technical documents were difficult for RAG systems to interpret accurately.