Chert: AI agent for data collection
Chert is a multimodal AI agent that transforms how users collect and interface with their data. It can take in unstructured voice recordings, images, videos, text, and GIS data, organize them into the right spreadsheet cells, answer questions about your database, and carry out actions or modifications, all through voice commands. This can be useful in situations ranging from field-based geophysics surveys to scheduling appointments and monitoring inventories. Overall, Chert offers a novel, customizable, and intuitive alternative to traditional digital forms and surveys.
Problem
Currently, industry data collection struggles with the inefficiencies of digital form and survey entry.
Workers in construction, field surveying, and other field based environments must juggle multiple tools, such as GPS devices, measurement equipment, and cameras, while simultaneously trying to document their observations by filling out forms. Construction punchlisting demonstrates this challenge. On larger sites, surveyors need to take GPS coordinates, take photos, and fill out digital forms while having their hands full with tools and materials. This fragmented approach leads to reduced productivity and inaccurate data capture.
Another example of this is from Cultural Resource Management, which borrows most of its spatial technology from construction. As someone who has personally worked with Sensys GIS data logging for a geophysics survey, I understand the struggles of recording data through forms while handling a GPS pole. In fact, it takes two people, one to balance the GPS pole and the other for data entry, to do something that can be simplified to one through a better interface than forms.
Additionally, users in retail and manufacturing environments suffer similar problems. Point of sale softwares, such as Square, Clover, or Shopify, currently require users to enter and monitor inventory manually through form entry. Analytics are also derived through a dashboard page, which are often complicated and unintuitive for users. This often leads to users using spreadsheets to manually keep track of inventory and bill of materials (BOS) data, thereby leading to even more confusion and losses.
Existing Solutions
Existing industry data collection softwares that connect with database softwares rely heavily on digital form entry.
For field-based data collection, solutions such as Survey123 or Fulcrum are forms that are built on the computer and deployed offline to mobile/tablet devices on the field for use. While solutions such as Fulcrum’s Audio Fastfill and Datch have begun integrating voice-based survey autofill, their interface is still preliminary and difficult to use (in fact, Fulcrum’s record, sign, and upload interface defeats the purpose of hands-free voice-based data entry).
For commercial and retail applications, point-of-sale softwares such as Square, Clover, or Shopify also require storeowners, barbers, or manufacturers to enter inventory and schedule appointments through manual form entry.
Our Solution
Our solution adds a reasoning layer between raw data collection and database entry in the form of an AI agent that understands, responds, and enters data based on the input query in the context of the environment and project. It is able to answer questions, provide analytics, and intelligently map unstructured multimodal data to structured database fields, whether that is cells on a spreadsheet or columns on a POS.
Users begin by importing their data tables through simple drag-and-drop exported CSV uploads or connection strings for databases and POS systems. Our system learns the database context through sampled data, extracted database schema, and user annotations on data fields.
The core innovation lies in our multimodal interface. After learning the database context, users can naturally speak their observations, capture contextual images or videos, and record GIS coordinates in an unstructured way. A reasoning layer between the user inputs and datatable automatically populates the appropriate fields in real-time based on the unstructured inputs and contextual understanding.
For users with relational databases, the platform supports relational mapping between multiple datasets. Users maintain complete control over export formats, whether JSON, CSV, or other common standards, since we focus purely on optimizing data collection rather than downstream usage. Users can also easily reconstruct relational mappings after exporting through the Excel Power Query tutorials we provide.
Conclusion
Chert addresses a fundamental inefficiency in how field data is captured and structured across multiple industries. This extends beyond just construction, preservation, and retail into any industry that requires incremental data collection through forms, such as museums, retail, or warehousing. The combination of a proven market demand, clear differentiation, and founder-market fit creates a favorable landscape for scaling and growth.