Research into technology that supports human agency.
We study the questions behind useful technology: what people need, what a system changes, and how to tell whether it helps. We turn those questions into careful experiments, with methods others can examine and limits we make clear.
Technology should expand human agency and recede when its work is done.
Our research begins with people: their time, their energy, and what gives their lives meaning.
In a world that keeps moving, we believe technology can help lighten the load and create space to pause, find clarity, and move with intention.
For us, human agency means being able to understand our options, make informed choices, and act on what matters to us. Technology should strengthen that ability while keeping people’s judgment and direction at the center.
That is what alignment means to us: not only whether a system works, but whether it supports the life a person wants to lead.
Our measure of progress: More capacity to act. More freedom to choose. More room to be yourself.
A higher bar for useful intelligence.
Four commitments guide what we choose to study, build, and bring into the world.
- Amplify human capability
- Respect attention and intention
- Build for the physical world
- Earn trust through evidence
Questions worth staying with.
These products are in development. Their names are working names.
NutriSpade
Flagship research · Early research · no outcome results yet
NutriSpade is our central research program exploring how technology could help people better understand what they eat and drink.
- Scope
- We are studying ways to make everyday nutrition information easier to understand and use.
- Evaluation plan
- We plan to assess accuracy, usefulness, and how clearly uncertainty is communicated.
- Limitations
- This is early research, not an available product. No NutriSpade outcome results are presented yet. Nutritional accuracy and usefulness remain to be established. No medical benefit is claimed.
SpadeBorne
Exploratory work · Exploratory · evaluation approach planned
A visual AI assistant that could help people understand a situation, weigh a decision, and act in the physical world.
- Scope
- Explore assistance that responds to a person's surroundings while leaving decisions and control with that person. The system and its evaluation are research directions, not an available product.
- Evaluation plan
- Task Explanation Reward will test whether explanations reflect the evidence and help with comprehension and task performance. The existing Explanation Reward Benchmark is an initial reference; its model reader results do not validate this broader program or its planned VR tasks.
- Limitations
- Context awareness, explanation faithfulness, human comprehension, and safe practical use remain to be demonstrated. A persuasive answer is not enough.
Related research, not product validation
Explanation Reward BenchmarkDJ Spadey
Exploratory work · Exploratory · evaluation approach proposed
Creative tools that follow the music. Explore video edits, lighting, transitions, and effects guided by rhythm and the creator's intent.
- Scope
- Study how an AI system might coordinate sound and visuals while letting a creator inspect, change, and refine its choices.
- Evaluation plan
- Art Explanation Reward is a proposed approach for testing whether explanations reflect the audio and visual evidence and help creators evaluate and refine the result.
- Limitations
- No outcome study is presented. Creative usefulness, reliable synchronization, and rights to music and visual material need separate evaluation. This is not an available commercial system.
Research & open source
Connection Dynamics
- Claim
- Early reciprocal interaction can add signal when identifying online connections that endure.
- Status
- Measured · research caveats apply
- Evidence
- A longitudinal cohort held apart for evaluation produced +4.48 ROC AUC points, 11.2× top 1% lift, and a 90 to 180 day outcome window.
- Boundary
- Ranking is not certainty. The results do not establish causality, compatibility in person, or an individual's future. It does not validate GatherLingo or Hello Again Assistant.
- Next validation
- The next question is whether these interaction signals transfer to relationships studied independently in person. That requires new evidence; the online benchmark cannot answer it on its own.
Open the research experience · View the repositoryExplanation Reward Benchmark
- Claim
- Test whether an explanation helps with a new task, not just whether it sounds persuasive.
- Status
- Provisional · human validation pending
- Evidence
- In a draft study before the corpus was frozen, structured explanations improved accuracy on new tasks by 12.5 percentage points over prose, with a +2.1 to +22.9 bootstrap interval.
- Boundary
- The corpus is a draft and the reader is an LLM proxy. Validation by human graders is pending, so no human comprehension claim is made.
- Next validation
- Complete blinded review by human graders and packet verification by two people before freezing a release. Then test whether the same representation effects help people with different knowledge and decision contexts.
Read the branch report · View the source branchLive Caption Bridge
- Claim
- Test whether bilingual captions can help people follow a shared event without taking attention away from it.
- Status
- Alpha · field validation pending
- Evidence
- A Windows English/Vietnamese alpha is implemented with 19 automated tests and Electron end to end validation.
- Boundary
- Venue accuracy, latency, audio hardware, two hour stability, and two rehearsal gates remain unverified. It is not validated for field or wedding use.
- Next validation
- Verify translation quality, latency, mixer/USB input, displays, network recovery, and two hour stability on the actual event setup. Complete two realistic venue rehearsals and retain a rehearsed fallback before relying on it.
View the platform branch · Inspect the alpha releaseFace ID Kit
- Claim
- Recognition assistance should keep enrollment, data, and uncertainty under human control.
- Status
- Implemented toolkit · application validation pending
- Evidence
- A locally running Python and Android toolkit with explicit enrollment and deletion, encrypted embedding storage, rejection of unknown matches, calibration separation, and synthetic implementation checks.
- Boundary
- Intended for voluntarily enrolled identities, not covert identification, authentication, or liveness assurance. Default thresholds are uncalibrated and scores are not probabilities. Human benefit, recognition across populations, and target device performance remain validation needs for each application. Python embedding encryption does not encrypt all identity metadata.
- Next validation
- Evaluate a specific, consent based assistive workflow with independent captures, representative participants, and real hardware. Measure mistaken matches and practical benefit; verify consent and retention controls before making an empowerment claim.
View the reviewed source · Read calibration & architecture · Inspect the Android modulesCinematic Web Framework
Open source framework
An open source React and Motion framework for building websites with AI collaborators. Editable components, project context, and quality checks keep the work inspectable. This homepage uses it.
A working example is not proof of developer productivity. Human design judgment, accessibility review, and testing are still needed.
Open the framework demo · View the repositoryOne company. Two connected roles.
EvoLovE is the research division of Spade Sense. We study the questions and evidence behind the technology. Spade Sense brings that work into product development, where each tool must be tested for its own purpose and setting.
- Spade Sense
- Our company brand and home for product development. Research informs the tools; each product needs its own testing before release.
- EvoLovE
- The research division. We study methods, examine evidence, and make the limits clear. Product development raises new questions for that work.
Explore Spade Sense productsTeam
Our research is led by Triet Pham, whose work spans applied AI, healthcare data, and system evaluation. The questions we study stay grounded in the settings where technology will be used.
Triet Pham, Founder · AI systems and applied research.
LinkedIn · GitHubProgress should leave people more capable.
More autonomous. More intentional. More able to shape the world beyond the screen.
