LatSpace Parser
Takes a factory’s messy monthly Excel sheet and maps its columns onto a standard parameter registry, flagging what it isn’t sure about.
- Kind
- Web service with human review
- Year
- 2026
- Built with
- FastAPI
- Next.js
- openpyxl
- Groq
- Llama 3.3 70B
- SQLite
- 1 more
The question
How do you get a dozen factories’ differently laid-out spreadsheets into one standard shape without trusting a model with the numbers?
The model only reads the column headers. Plain code reads the values, and a person settles the doubtful columns.
Why it exists
Factories send monthly operational spreadsheets in inconsistent formats. LatSpace, an industrial ESG platform, tracks parameters such as coal consumption, steam generation and CO₂ across assets like boilers, turbines and kilns, so each column has to be mapped to a standard parameter first.
How it works
From a messy workbook to standard parameters
- 1
Workbook
An uploaded monthly Excel file, up to 25 MB.
- 2
Header
The header row is found as the first row that is mostly text.
- 3
Mapping
The model maps each column header to a registry parameter and asset, with a confidence score.
- 4
Values
Cell values are parsed by code in chunks of 500 rows, then checked against domain rules.
- 5
Review
Doubtful columns reach a person, who approves, rejects or overrides them.
- 6
Export
JSON, CSV, or a CSV of what could not be mapped.
Decisions
The model maps names, code reads numbers
Keeping cell values away from the model means it cannot change a figure. It only decides what a column is called.
One call per sheet
A single request maps every header on a sheet, and a second pass runs only when there are doubtful mappings, which keeps cost and delay low.
A person can overrule the model
An override sets the confidence to 1.0 and records that a reviewer made the change, and decisions are kept in their own table.
Evidence
- Built
Five example workbooks (clean, messy, multi-asset, edge cases and sparse) are served for a try-out.
Built with
Frameworks and libraries
- FastAPI
- Next.js
- openpyxl
Data and storage
- SQLite
AI and models
- Groq
- Llama 3.3 70B
Infrastructure and services
- Docker