JevMade Sign in
← Back to guides

JevMade field notes / Early implementation case study

Put a fast routing model beside an AI planner

Procurement Mind races a Jev message router against its usual AI planner while keeping purchasing decisions with trusted rules and people.

Original by Alvar LaignaAgent workflows

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits

“System One Next to the LLM: First Week in a Procurement Assistant” by Alvar Laigna. Read the original source.

This expanded guide is an AI-narrated adaptation prepared by JevMade. It expands the source’s essential ideas, examples and caveats in JevMade’s own words and is not a word-for-word reading. The synthetic voice does not imitate the author or imply their endorsement.

Our summary

Procurement Mind helps Estonian public buyers prepare documents and supplier lists. Alvar Laigna describes adding Jev for small decisions: where a message belongs, which passages answer a question, and which purchasing codes fit. The app still uses a language model to plan and draft.

For message routing, both models start together. A sufficiently certain Jev answer can take the fast path; otherwise the planner continues. Some planner calls finish anyway to compare their choices. Each decision can be active, checked without acting, or switched off, and direct names are replaced before text leaves.

The first week covered only sixty decisions, mostly internal testing, with eleven confident routing agreements. Agreement with the planner is not proof that either route is correct. Faster routing does not establish better procurement documents, and stronger Estonian support and an EU service region remained open concerns.

Key takeaways

  1. Run a fast router beside the planner rather than adding another wait in front of it.
  2. Keep comparing decisions after launch and make each new path easy to disable.
  3. Keep legal and workflow authority with trusted rules and people, not an AI probability.

This is an early public-purchasing trial, not an established result in regular use. Cost and timing figures are the author's early observations. Removing direct names before sending text does not establish complete privacy protection. No implementation, timing result or claim about purchasing quality was independently tested.

alvarlaigna.com · Original published

Read the original guide Opens the author’s site in a new tab.

Keep this for later

Sign in to bookmark experiments, guides and videos, and keep notes only you can see.

Continue to sign in

We’ll bring you back to this listing.