Nearu

**Product and concrete problem.** Nearu is developing infrastructure for a specific weakness in current AI agents and embodied systems: they can generate fluent language, but they generally do not perceive the human context that determines whether an interaction is reassuring, confusing, unsafe, or about to fail. The company positions its Soul Engine™ and NearuVibe™ products between a human-facing application and the application’s underlying language model. That layer is intended to recognize emotional signals, preserve context across sessions, and shape the agent’s response. The practical problem spans several high-stakes settings: an onboarding agent that misses a new employee’s anxiety, a support system that escalates a frustrated customer, a healthcare assistant that fails to notice distress, a tutor that keeps teaching after a learner is lost, or an in-car assistant that ignores driver stress. Nearu’s product thesis is that an AI system needs a structured perception and relationship layer, not only a larger language model, if it is expected to operate reliably around people.

**Core technology and how it works.** NearuVibe™ is described as a multimodal emotion engine that intercepts audio and video before the language model responds. The official product site separates three signals: acoustic analysis of voice prosody such as pitch, pace, volume, and tremor; semantic analysis of the words and their emotional meaning, including contradiction or masking; and facial-expression analysis based on a lightweight visual emotion-recognition model. Nearu says its facial model was trained on approximately 450,000 faces from AffectNet, and that the channels are combined with confidence weighting rather than treated as interchangeable evidence. The output is a structured emotional signal delivered through an API, including labels, confidence, trend, and an evidence summary. The company advertises 50–100 millisecond acoustic-frame processing and an approximately 1.6-second live-session detection experience, but these are company-reported product metrics rather than independently benchmarked results. The broader Soul Engine combines that perception output with memory, identity, and a personality matrix, while remaining model-agnostic across GPT, Claude, Gemini, or local and on-premise models.

**Market, customers, and go-to-market.** Nearu is pursuing a B2B infrastructure model rather than selling only a consumer companion. Its public materials ask for design partners, robotics OEMs, and teams building human-centered AI, and identify enterprise employee experience, customer support, healthcare, education, sales, automotive assistants, and companion robotics as application areas. The strongest initial wedge is likely an API or SDK embedded by a product team that already owns an agent, avatar, robot, or voice interface and needs a real-time behavioral signal without retraining its primary model. This creates a potentially attractive horizontal position, because the same perception and memory services could be reused across verticals, but it also leaves Nearu dependent on integration partners and on customers deciding that emotion-aware behavior improves measurable outcomes. Public sources do not establish pricing, signed design partners, recurring revenue, conversion rates, or a named enterprise customer. Nearu’s own demonstrations show a live session in which tone, words, and facial cues are displayed separately before an adaptive response, which is useful as a product proof point but is not customer traction.

**Traction, funding, and third-party validation.** The Startup Nation Finder profile lists Nearu as founded in November 2025 by Noa Shapiro and Vladimir Kolesnikov, with 1–10 employees and a pre-funding, R&D-stage status. IVC lists the company as established in 2025, with four employees, an enterprise-software and infrastructure classification, and a statement that it was seeking $2 million as of May 2026. Nearu’s LinkedIn page says the company is backed by NVIDIA Inception, Google for Startups, and AWS Activate; those memberships can indicate access to cloud, technical, and ecosystem resources, but they do not prove investment, revenue, or product-market fit. The official website exposes a live demonstration, REST-style emotion-analysis endpoint, API and WebSocket references, regional deployment options, and a privacy policy, giving more technical specificity than a concept-only landing page. The public record still lacks an independently audited accuracy study, a disclosed financing round, a patent list, a contract announcement, or customer outcomes. The diligence conclusion is therefore early technical validation and ecosystem participation, not commercial validation.

**Founders and team.** Nearu’s official team page identifies Noa Shapiro as founder and CEO, with more than 12 years in business development, marketing, and go-to-market roles and a graduate business degree in progress at Reichman University. Her public LinkedIn profile describes the company’s core as a perception API that fuses voice prosody, semantic context, and facial expressions before an LLM responds. Co-founder and CTO Vladimir Kolesnikov is presented as an R&D executive and AI strategist with more than a decade of experience; Nearu credits him with serving as Head of R&D at Elfi-Tech, co-founding an AI lab later acquired by Zepp Health, and building AI solutions for Samsung and Xiaomi. His own technical profile emphasizes hands-on architecture, LLM and retrieval-augmented generation systems, agentic workflows, and regulated-product experience. Doron Pryluk is listed as strategic advisor and product evangelist, with prior operating roles at Quack AI and Colleen AI. This is a complementary commercial and technical founding shape, but the public team is small and the company has not yet disclosed a larger research bench, clinical advisors, or a published scientific leadership group.

**Competitive dynamics and edge.** Nearu competes at the intersection of emotion recognition, conversational AI, digital humans, and robotics infrastructure. Hume AI and audEERING compete on speech and multimodal emotion understanding; Realeyes and Smart Eye compete on camera-based attention and affective analytics; Soul Machines and Inworld AI compete on persistent, embodied, or character-driven agents; and the platform companies behind major voice assistants can incorporate similar capabilities into their own stacks. Nearu’s claimed edge is architectural: its 3-channel acoustic, semantic, and facial fusion is intended to capture mismatches that a text-only model misses, its structured API can feed any selected LLM, and its memory and identity layer is intended to compound relationship context over time. That model-agnostic position could be valuable to robotics OEMs that do not want their social behavior layer locked to one model provider. The counterargument is that large model vendors can add emotion classification, memory, and multimodal context as platform features, while specialist competitors may have stronger peer-reviewed datasets, enterprise distribution, or validated accuracy. Nearu’s defensibility therefore depends on measured performance across languages and cultures, low-latency deployment, privacy-preserving data handling, and evidence that the behavioral layer produces better operational outcomes.

**Defense, security, and resilience relevance.** Nearu has no public evidence of defense contracts, military deployment, security accreditation, or use in operational command systems, so its core technology should not be described as fielded dual-use defense capability. Its strategic relevance is an adjacent AI-infrastructure and human-machine-interface thesis. Nearu’s multimodal perception could eventually support resilient remote care, operator training, emergency communications, driver-safety systems, or robots working around stressed and vulnerable people; its privacy policy also describes regional cloud deployment, on-premise options, deletion controls, and restrictions on using customer data to train shared models. In a future defense or critical-infrastructure setting, emotion and behavioral context might help an interface recognize overload or confusion and route a decision to a human, but that is a product hypothesis requiring validation, not a current capability claim. The immediate resilience case is civilian: safer interaction with healthcare, education, automotive, and service systems where missed human signals can create harm or service failure. This record consequently sets dual_use to false while preserving Nearu as a potentially relevant enabling layer for future human-centered robotics and trusted AI.

**Growth stage, trajectory, and diligence risks.** Nearu is best classified as early stage: it was founded in late 2025, is described by Finder as R&D and pre-funding, is seeking a Seed round according to its website and IVC, and has not publicly disclosed revenue or a customer deployment. Its trajectory could become meaningful if it converts the live demo and API into paid design partnerships with robotics OEMs, healthcare or automotive platforms, and enterprise agent vendors, while publishing reproducible accuracy, latency, calibration, and bias results. The main diligence risks are (1) scientific validity and cultural generalization of emotion inference, especially when facial and vocal signals conflict; (2) privacy, biometric-data, employment, healthcare, and mental-health regulation; (3) the absence of independent performance evidence and disclosed customer outcomes; (4) intense competition from model vendors and better-funded affective-computing specialists; (5) dependency on camera, microphone, transcription, and LLM infrastructure in real deployments; and (6) limited financing, headcount, and operational history. Nearu merits monitoring for its concrete API and unusually explicit multimodal architecture, but priority should rise only after independent validation, paid design partners, transparent data-governance reviews, and evidence that its emotional layer improves safety or retention rather than merely making demonstrations feel more personable.