Provenance first
Claims remain connected to synthetic source IDs, frames, events, and generation records.
Independent Multimodal AI Research
From observations to grounded reasoning.
Visual research notes and interactive demos that connect video, text, audio, and sensor observations to temporal evidence, uncertainty, provenance, and explainable conclusions.
Video Text Audio Sensors
temporal evidence graph
HYPOTHESIS UPDATEsupport 0.76uncertainty ↓ 0.31
Grounded answer →
Research principles
Make the reasoning inspectable.
Each prototype keeps observations, assumptions, uncertainty, and limits visible.
Reasoning across time
Timelines reveal when support rises, failure occurs, or a contradiction changes the state.
Uncertainty made visible
Confidence and entropy are shown as model outputs with explicit assumptions—not facts.
Multimodal evidence
Video, text, audio, and sensors contribute distinct, traceable observations.
Reproducible synthetic demos
Compact deterministic scenarios make expected behavior testable without real people or cases.
Public research origins
Established prior work is separated from PersenseAI implementation choices and limitations.
Featured projects
Four views of temporal reasoning.
A sequence from dense progress signals, to provenance-backed claims and evidence utility, to bounded evidence-tool orchestration.
Unified research pipeline
One traceable path through multimodal evidence.
01Multimodal
Inputs
→Inputs
02Observations
→03Temporal
Evidence Graph
→Evidence Graph
04Utility +
Uncertainty
→Uncertainty
05Evidence
Orchestration Agent
→Orchestration Agent
06Grounded Answer or
Inconclusive Stop
Inconclusive Stop
Agent Mode
Choose the next evidence action. Know when to stop.
A deterministic utility-driven policy ranks only registered tools, updates a typed state after each call, and stops with citations or an explicit inconclusive result. It is bounded by tools, steps, cost, and stop rules; no language model is used for planning.
Open the bounded agent demoSTRUCTURED TRACE PREVIEW01 inspect visual observation +0.5102 update evidence graph +1.9603 inspect contradiction +1.3104 run counterfactual +1.11→ grounded answer or inconclusive stop
Live static mini-demo
Which evidence changed the conclusion?
Client-side synthetic data. No backend, model call, external API, or analytics.
BOUNDED HYPOTHESIS
Support
Uncertainty
COUNTERFACTUAL
NEXT USEFUL EVIDENCE
Deterministic synthetic research prototype. Utility is an explanatory model—not truth, legal weight, or forensic validity.
Research notes
Origins and limits stay beside the demo.
Every demo begins with public research or an independently documented question. Established concepts are separated from implementation choices; limitations are explicit; provenance is part of the interface, not an afterthought.
Read the research-origins note →About PersenseAI
An independent research space for multimodal reasoning.
PersenseAI explores how observations become grounded explanations across time, evidence, uncertainty, and provenance. The work is presented through public notes and compact synthetic demos—not production or forensic systems.
Working on temporal reasoning, multimodal evidence, evaluation, or provenance? I’d be glad to compare notes.
hello@persenseai.com