iFLYTEK: speech, models and hardware in everyday life

AI becomes visible in the work someone needs to do

The visit to iFLYTEK took us through a showroom of artificial intelligence applications. The tour brought together a company presentation, translation, teaching tools and support for medical work. It helps shift the question: beyond learning about the model, what matters is understanding the activity it supports and the experience delivered to the person using it.

In education, the demonstrations included tools for representing mathematical concepts, recording lessons and supporting the marking of assignments. In healthcare, the presentation covered an assistant integrated into a doctor’s work. Translation was also explored as a way to connect people speaking different languages. These examples give concrete shape to the discussion of applied AI.

One takeaway for founders is that an AI product becomes meaningful when it connects technical capability, its context of use and distribution. That connection is where a demonstration can start to become a business.

A journey that begins at university and spans different technology cycles

iFLYTEK was founded in 1999. Its institutional history describes its origins in a group of university classmates, the development of speech research and its listing on the Shenzhen Stock Exchange in 2008, under the code 002230. The open platform became part of that journey in 2010. The company was building language solutions long before the current rise of generative AI. Official iFLYTEK history.

The University of Science and Technology of China (USTC) identifies Liu Qingfeng as a founder and alumnus in an account of his participation in an academic event. The cognitive intelligence laboratory jointly maintained by iFLYTEK and USTC also recounts the role of Professor Wang Renhua, who led the company’s beginnings alongside Liu and their university colleagues. This history connects education, research and entrepreneurship. USTC institutional account and History presented by the joint iFLYTEK and USTC laboratory.

For founders, this history opens a discussion about continuity. Expertise developed around a specific problem can find new applications as computing capacity, the market and the interface change. Sustaining research and revisiting how to deliver value are efforts that can move together.

Speech and language form an infrastructure for interaction

A conversation with a machine involves several stages: capturing sound, recognizing what was said, interpreting the request, retrieving information and producing a response. Translation also requires carrying the meaning into another language. The experience seems simple when it works, but it depends on the quality of the entire chain.

iFLYTEK’s public international catalog includes speech recognition, text-to-speech, translation, pronunciation assessment and recognition of text in images. These capabilities can be integrated into third-party applications. Rather than requiring every company to develop all the components, the offering makes it possible to combine tools around a specific use. International developer platform.

For entrepreneurs, it is useful to observe where language reduces friction. Someone might need to look up information while working with their hands, record a meeting or communicate with a person from another country. The value lies in the task completed, the understanding preserved and the time no longer spent on intermediate steps.

In the classroom, the interface brings an idea closer to understanding

One part of the visit presented digital whiteboard tools and mathematical visualization. The explanation connected drawings, shapes and three-dimensional representations to the possibility of understanding spatial concepts more clearly. This matters because part of learning depends on how an idea is presented, explored and revisited.

Lesson recording was also discussed, allowing students to return to the content later. The presentation connected what happens on the board and in the explanation to the material students can consult. It also addressed marking assignments and helping teachers understand learning difficulties.

For public context, iFLYTEK brought its Spark Education LLM and digital education solutions to the international education conference held in Hangzhou from May 11 to 13, 2026. Its offering is organized around teaching, learning and educational management. World Digital Education Conference 2026.

The reach of digital infrastructure helps put these demonstrations in perspective. In September 2026, China Education Television (CETV), attributing the figure to China’s Ministry of Education, reported that 99.1% of primary and secondary school classes were covered by interactive multimedia resources. This indicator describes teaching infrastructure: it does not mean that 99.1% of schools use artificial intelligence, nor does it measure iFLYTEK’s presence. Educational digitalization context from CETV.

There is also an AI-specific indicator with a different scope. According to an account published on the Ministry of Education’s portal, coverage of artificial intelligence applications in Beijing’s educational institutions reached 87.7% by the end of 2025. The figure encompasses different types of institutions in the city and cannot be generalized to all of China. Having equipment, adopting AI applications and improving learning are different stages, each requiring its own assessment criteria. Report on AI adoption in Beijing.

For an edtech company, the opportunity is to investigate the whole process: what difficulty a student faces, how the teacher identifies it, what intervention might help and how the result will be observed. Automating an activity only makes educational sense when it improves some part of that relationship.

Giving teachers time back must come with useful feedback

The presentation mentioned analysis and marking tools to support teachers. The business question that follows is concrete: does the tool return information that helps someone decide what to do next? A lengthy report can increase the workload if it does not show where to act. A clear signal can help a teacher prepare an intervention or a different explanation.

A takeaway for Brazilian founders is to design the offering around the professional user. In education, that includes recognizing the school’s conditions, the time available and how activities are already prepared and assessed. The model is part of the solution, but the experience also depends on integration, language and the ability to review the output.

A suggested pilot metric could combine time spent on an activity, usefulness as perceived by the teacher and the quality of the subsequent intervention. The comparison should consider the same type of task and allow users to correct the result. Assessment needs to be adapted to the context.

The assistant enters the doctor’s workflow

The healthcare presentation described support for doctors within their working environment. It placed the system on a computer used by the professional and connected the technology to assistance in everyday practice. The central point is the integration of information and human decision-making.

The company’s public communication about WAIC 2026 describes its clinical decision support system as a tool that provides evidence to doctors. The text itself distinguishes that support from the clinical decision, which remains with the professional. That formulation is the reference used here to contextualize the demonstration. Official WAIC 2026 account.

The business application is to examine the chain of responsibility. A tool can organize information or reduce administrative steps, while the specialist still needs to assess what it produces. For a startup, understanding who uses it, who validates it and who is accountable for the result is part of defining the product from the outset.

Spark provides a foundation for different applications

In the company presentation, Spark was described as a large language model that can provide a foundation for other applications. The discussion also connected that foundation to Chinese computing infrastructure. The interest for the trip is in observing the connection between research, models and solutions that reach different sectors.

The public Spark LLM offering includes document-based knowledge retrieval, coding support and collaboration between agents. In July 2026, the company’s education division announced updates to the Spark X2 family, including versions serving different performance and multimodal needs. These are the company’s own descriptions. Public Spark LLM offering and Official July 2026 update.

The same July communication connects this development to a partnership with Huawei for the Feixing No. 1 computing platform, announced in 2023. The company presents the development of Chinese infrastructure as part of its technology strategy. This context provides a public reference for the topic of computing autonomy mentioned during the visit. Official July 2026 update.

For a startup, the useful distinction is between general capability and solving a problem. A foundation can serve many uses, but a product needs to choose which data it accesses, what task it performs, how it responds to an error and what it delivers to the customer. Specialization takes place in that design, beyond the choice of model.

The figures help show the scale of an applied AI business

The summary of the 2025 annual report, released on April 29, 2026, reports consolidated revenue of approximately RMB 27.1 billion, up 16.12% from 2024. Education accounted for RMB 8.97 billion. These are 2025 figures for the company’s business across different areas. Summary of the 2025 annual report.

The company’s size helps reveal the importance of implementation. An AI company’s financial results may bring together software, services, equipment and projects across sectors. To assess an opportunity, it is necessary to identify where revenue is generated and what work is required to sustain delivery.

The open platform turns technical capability into distribution

A distribution strategy can begin when other teams are able to build on the technology. iFLYTEK’s international platform offers access through an account, console and application identifier, as well as programs for technology and channel partners. This design brings technical components closer to companies that already understand a market or an operation. International developer platform.

For founders, this broadens the meaning of distribution. It can take the form of an integration, a device from another brand, a partner implementing the solution or a developer experimenting with an API. Each route requires its own adoption experience: documentation, testing, support and a clear reason to incorporate the capability.

A question for founders is how to create a partner’s first win. What task can they accomplish with the technology in a short time? How can they demonstrate value to their own customer? A platform that enables this progress creates conditions for its capabilities to spread across products. This is an interpretation of the offering, not a measurement of its conversion rate.

The solution can also arrive as a service, package or project

During the questions, the offering was also explained in terms of a service, package or project, configured according to demand. The discussion broadens the view from an isolated device to the work required to deliver a solution. A customer may need a ready-to-use function or an application integrated into its operations.

The closing discussion also addressed differences between versions intended for the Chinese and international markets. The explanation connected adaptation of the offering to the demands of each market. For entrepreneurs, this helps identify what needs to change when a product starts serving other contexts.

The takeaway for entrepreneurs is to separate what can be standardized from what requires a specific implementation. This helps define how the solution is presented, which capabilities are involved in delivery and what follow-up will be needed afterward. Understanding these differences allows a better assessment of the effort involved in serving each type of customer.

A pilot can demonstrate value in a small setting, while organizational adoption requires integrating processes and people. For a startup, the question is how to preserve quality when moving from an initial experience to repeatable delivery. Commercial learning includes understanding the work between a demonstration and continued use.

The device can be the entry point for AI

The company’s public offering combines software with dedicated devices. AINOTE 2 brings together writing on an e-ink screen, recording, transcription, translation and meeting summaries. The format places these functions in an object associated with taking notes and following conversations. AINOTE 2.

At MWC 2026, the company presented glasses with translation capabilities and the AI Interpret Mic for communication in meetings. These announcements show different ways to bring language functions into real situations, beyond a mobile app. They illustrate how AI capabilities can enter everyday life through different interfaces. iFLYTEK’s presentation at MWC 2026.

The takeaway for entrepreneurs is to start with behavior. A device can make capture, access and continuity easier, but it also adds manufacturing, maintenance and support decisions. The question is what advantage the physical form offers for that activity and how users experience the advantage in daily life.

The experience depends on a combination of components

During the questions, features of a transparent screen and a whiteboard were compared. The presentation distinguished camera functions, including recording the scene and using visual cues to identify who is speaking during an interaction. This connects interface design to the activity it needs to support.

The AIUI platform offers development interfaces, WebSocket connectivity and hardware modules. Its catalog covers voice interaction, multimodal capabilities, noise reduction and scenarios with connectivity or offline operation. It also presents integration with third-party models, agents and knowledge bases. AIUI interaction platform.

This combination helps explain why the product is more than the model. Perceived quality depends on capture, response time, the content accessed and how the response reaches the person. In environments with movement or noise, interface components can determine whether the function is actually used.

One approach to a pilot is to test the entire chain in real conditions. A result obtained in a quiet room may behave differently at a service counter or during an operation. It is useful to measure the task completed, record where it fails and design an alternative that allows work to continue.

When AI gains a body, execution becomes part of the product

As additional context, the company presented robotics initiatives with Lingdong General at WAIC 2026. The announcement identifies that company as the manufacturer of the LDB01 and LDT01 robots and describes its collaboration with iFLYTEK RESEARCH on a multimodal model connecting understanding of the environment with action. The division of roles matters when understanding the ecosystem. Official WAIC 2026 account.

The shift to what is called physical AI adds questions of perception, movement and operation. Understanding an instruction is one stage. Turning it into useful action requires observing the environment, handling variation and assessing the outcome. A solution also needs to reach the workplace, be maintained and keep functioning over time.

For the NewHack community, this opens opportunities across the chain. A team could work on the interface, data collection, task integration or operational monitoring. The suggested starting point is an observable problem with clearly defined execution conditions and responsibility.

Choosing a deployment also means choosing how the product will operate

The public Spark LLM page describes cloud API options, on-premises deployment, collaborative training and solutions combining hardware and software. These are different commercial approaches serving integration and data-control needs. The specific configuration depends on the product and the deployment agreement. Public Spark LLM offering.

The takeaway for founders is to turn architecture into business questions: where must the data remain, who can access it, what response time is acceptable and how does the customer operate if a connection fails? The answers may change product design, the support required and the cost of serving that market.

It is also useful to separate what the team needs to develop itself from what can be integrated. Having a distinct product identity does not require building every component. The strategic work is to understand dependencies and create an experience whose quality can be monitored and improved.

Translation becomes more useful when it understands the travel and service context

The visit explored communication between languages, a need the group itself experiences while traveling through China. The application becomes tangible when people can follow a conversation, ask a question and confirm the meaning of an answer. The experience depends on both the technology and the context in which it is used.

In an official account of the December 2025 China International Travel Mart, published in January 2026, iFLYTEK presented dual-screen translators, translation earbuds and virtual characters for tourism services. Examples include hotel reception, visitor guidance and accompanying groups. Public translation applications in tourism.

For Brazilian entrepreneurs, the discussion points toward localization. A capability may need to understand a sector’s vocabulary, regional expressions and how people actually ask for help. Choosing a specific context makes it possible to assess those details and create an offering closer to the user’s needs.

Turning the learning into a pilot that can be evaluated

To design a pilot, it is useful to organize the opportunity around the work to be done. A promising application brings together a defined user, a recurring difficulty and a way to observe whether there has been improvement. This design enables learning before expanding scope and helps bring technology, partners and customers together.

The opportunity lies in connecting capability, context and adoption

iFLYTEK broadens the trip’s perspective on Chinese innovation. Its history connects university research and technological development. The examples presented bring that foundation closer to teaching, communication and medical work. The public offering shows how models, APIs, partners and devices can provide different routes to the user.

For NewHack, the lesson is to view the solution as a chain of relationships. Those who understand the problem, develop the capability, implement it, use it and validate it need to find a way to work together. Community and distribution matter in that connection because they help match capabilities with concrete opportunities.

The challenge for founders is to choose where they can create value and learn from use. Technology opens possibilities. A well-defined problem, verifiable delivery and a path to adoption help turn those possibilities into a business decision.