From the workbench
What we’re building.
What we’re learning.
A record of implementation decisions, practical research and questions we are testing as Anvayric develops.
Latest notes
Follow via RSSBusiness knowledge changes.
The application has to keep up.
Review, approval and versioning are part of an AI application—not housekeeping after the model.
When should a workflow
become an agent?
Rules, interpretation and tool use solve different parts of the problem.
From IMFIT to
applied AI engineering.
What the implementation proves, and what the market still needs to tell us.
Agents, connected data
and the smallest useful model.
The questions we want to test, the evidence we have and the results still to be earned.
Evidence before claims.
Implementation notes describe work we have inspected. Research articles link to their sources. Proposed experiments are labeled as proposals. Results will be published when they have been measured; a regular publishing schedule is not yet established.
Tell us what you’re interested inScoped AI engineering
Have a task in mind?
Let’s assess the fit.
Start with the work, the information it needs and the question you want to answer.