Evidence-guided decipherment
Early prototypeHow could verified clues about meaning help us investigate incomplete language evidence while keeping uncertainty visible?
Tests on constructed examples do not establish real-language decipherment.
AI-assisted products. Independent research.
Where insight takes shape.
Anavi develops thoughtful software and pursues independent research. Our current build is Library Twin—a digital counterpart to your physical library.
Begin withA question.
Open intoPossibility.
The current build
A library is more than a list.
The books you own. The places they belong. The context you want to keep.
In developmentYour collection.
Library Twin
Your collection.
Its connections.
A digital counterpart to your physical library—designed to connect books, editions, individual copies, locations, and the stories that make them yours.
Visit Library TwinKeep the particular copy.
A title, an edition, and your own copy are different things. The product is designed to preserve that distinction.
Find the thread.
Explore a collection through questions and relationships—not just a list of titles.
Carry the context forward.
Keep places, notes, and meaning connected to the books that matter to you.
One development decision, explained.An example of evidence-aware answers
“Not sure” should not become a wrong “no.”
A shortened title could return an incorrect “not owned.” The documented Library Twin repair preserved a possible owned match and made the uncertainty explicit.
Illustrated reconstruction of a recorded development case. Not a live app demonstration.
The question
“Do I own Moby-Dick?”
A shorter title than the stored test record.
Possible match
Keep the evidence in view.
The documented repair returned uncertainty with a possible owned match. Its link opened the relevant copy in the recorded development preview.
Based on a September 11, 2026 development record. Wording is reconstructed; these controls do not access the app or rerun its tests. The bounded repair does not establish general accuracy, current production behavior, or release readiness.
Independent research
Room for the next question.
Selected directions under investigation. Each has its own evidence and its own next step.
Mathematical reasoning
Experimental designHow can different models work together to compare explanations and identify what to investigate next?
The complete system and broad performance claims still need validation.
Research to implementation
Workflow developmentHow can decisions, evidence, and next steps stay intact when AI-assisted work passes from one session to the next?
An evolving workflow, not a system that builds apps on its own.
The person behind Anavi
I’m Ash.
Founder · AI product builder · Independent researcher
I’m building Anavi around useful software and questions worth following. My work connects AI-assisted product development with research that crosses conventional subject boundaries.
AI is a collaborator. Deciding what matters—and checking what is true—remains part of the work.
U.S. Army veteran. MBA student. Building Library Twin.
What are you curious about?
For questions about Anavi, get in touch by email—or follow Ash’s work on X.