01 / Accounting with agents
Auxiliar AI keeps the monthly close connected
How source data, specialist agents and professional decisions fit together across the monthly accounting cycle.
Read the technical study
Embedding Labs / Systems & engineering
Inside the systems.
The architecture and decisions behind five Embedding Labs systems.
Read the business stories
How we build
Four principles.
Shown in real systems.
- 01
Keep proposals separate from authorization.
A system prepares the work; a person with the authority approves what changes.
- 02
Keep the source, client and period with the result.
Every output stays traceable to the records and context it came from.
- 03
Test the workflow, not only the model.
Evaluation covers the normal path, the exceptions and the resulting changes.
- 04
Use retrieval for information and fine-tuning for behavior.
Facts come from current sources; training shapes how the system works.
02 / Driver planning
A clearer planning horizon for drivers
Availability estimation, deterministic assignment and human dispatch judgment form one planning workflow.
Read the technical study
03 / Talk to your data
Talk to your data
The engineering behind Talk to your data: format-aware document recovery, contextual extraction and managed vector indexing that prepare scattered files for retrieval-augmented generation (RAG).
Read the technical study
04 / Procurement taxonomy
A common language for purchasing
Reference retrieval, structured classification and a separate matching analysis prepare a four-level procurement crosswalk.
Read the technical study
05 / Legal claim response
From a claim file to a reviewable response
Structured extraction and a shared plan coordinate research, specialist sections and an attorney review surface.
Read the technical study
Carlos Molina's earlier banking and payments systems are on his own page.
Carlos Molina, track record