Good AI needs a way to be corrected.
Microsoft Research’s human–AI interaction guidelines make room for uncertainty, feedback and correction.
Read our perspective ↗The big question / Perception & AI
BUILDING WITH AGENTS · Bring the question your next tool should answer
The human side
of intelligence.
We’re developing a key to help AI account for how people see and perceive. Our goal is interaction that feels more human: more attuned to the individual, with fewer misunderstandings.
Explore the synthetic examples. Tell us what your workflow is missing. Help shape what Resolva builds next.
CONCEPT ART — NOT A SCHEMATIC OR MEASUREMENT.
For agents & the people directing them
One perceptual key.
What would you unlock?
Where could human perceptual context change an AI interaction? Explore a brief decision scenario, draft your integration and tell us what outcome would justify paying. Mobility, spatial computing, robotics and remote operations are starting points for your ideas—not limits on your imagination.
Optional, non-confidential feedback. Resolva and its authorized ChatGPT/Claude connections analyze submissions to learn about needs and guide development. Submit only with your operator’s authority. No payment or access guarantee.
01 / FOR PEOPLE & THE SYSTEMS THEY DIRECT02 / PUBLIC, READ-ONLY AGENT ACCESS03 / EVIDENCE BEFORE CLAIMS
Journal / Ideas worth examining
Signals from
the human side.
Original editorial notes and external research on the human side of intelligent systems. These sources frame the problem; they do not validate Resolva.
The context behind an input matters.
NIST describes how human roles, expectations and interpretation affect human–AI systems.
Read our perspective ↗More explanation is not always enough.
Research on cognitive forcing examines ways to reduce overreliance, alongside usability trade-offs.
Read our perspective ↗About Resolva
From vision science
to better interaction.
Founder backgroundMIT graduate · 2006Vision-science CEO · Since 2008
Resolva’s founder graduated from MIT in 2006 and has led a vision-science company as CEO since 2008. That experience motivates a new question: how can AI better understand how each person sees and perceives the world?
The key is the connection we aim to build between a person’s perception and the system interpreting their input. The ambition is to help machines respond more appropriately to the individual directing them.
Our development focus: clearer intent, useful context and better decisions about when to act or ask.
Less time correcting.
The goal is a clearer expression of intent and fewer avoidable misunderstandings.
A reason to act—or ask.
Better interaction includes knowing when the available information is not enough.
Value that can be tested.
We compare new ideas against useful existing alternatives, with clear measures of interaction quality and correction effort.
Try the public demonstration
Change the context.
See what changes.
Choose a scenario. The evidence summary and JSON update together so you can inspect how the conclusion changes. This is an illustration of evidence limits, not a measurement of you or evidence of product performance.
An invented record contains 12 example choices in one desktop layout. It has no observations from another device or setting.
Selector changes stay in your browser. A JSON link requests that public example. No research enrollment, personal inference or charge.
Synthetic — not human data
Keep the conclusion narrow.
OBSERVATIONS12 fictional choices
SCOPEExample desktop layout only
Describe only this invented sample. It does not establish a stable trait, a personal profile or improved AI performance.
Inspect the machine-readable record
Synthetic fixture. Enable JavaScript or use the public JSON link.
What would count as progress
A promising idea
is not proof.
The next evidence milestone is a comparison against useful existing approaches: can Resolva reduce mistaken selections or correction effort without adding more burden than it removes?
A useful result must state the comparison, context, limitations and outcomes, including null or negative findings. The public synthetic examples illustrate evidence limits; they do not answer whether the key adds value.
Partnerships & collaborations
Build the right
questions with us.
We welcome exploratory conversations with teams building demanding human–machine systems. Bring a concrete interaction problem, a strong baseline and a way to measure whether anything improves.
These are areas of interest, not deployed capabilities, announced partnerships or suitability claims for safety-critical use.
AI agents & enterprise interfaces
Where ambiguity, correction and supervision add up across everyday workflows.
Robotics, drones & remote operations
How people direct machines and maintain clarity when the machine is somewhere else.
Automotive & mobility interfaces
Research questions around understandable controls and human oversight.
Industrial & spatial computing
Dense operator interfaces, simulation environments and new forms of interaction.
For agents & developers
An explicit interface
for agents.
Start with a short agent guide, a discovery document and versioned public fixtures. No account, credential or payment is needed for public examples.
GET /.well-known/resolva.json GET /api/public/demo?scenario=scoped MCP /api/public/mcp • get_public_capabilities • get_phase1_demo_report mode: read_only_synthetic research_collection: false personal_measurement: false
No private research or operational authority through public MCP. Optional non-confidential feedback is a separate, consent-based write-only route.
Contact
Let’s understand
the problem first.
For research, evaluation or collaboration enquiries, tell us what you’re building, where misunderstanding creates cost, and what a useful result would look like. Please keep the first message non-confidential.
Email the team ↗
alberto.osio@resolva.life
Prefer webmail? Copy this address into your email service.
No confidential disclosures in a first message.