Inside Intercall’s real-time AI: built to assist interpreters, not replace them

For interpreters who depend on it daily, the feedback is unanimous: finally, a tool designed around how they actually work. The core idea is straightforward. The best interpreting happens when humans and machines collaborate, not when one replaces the other.

Real-time interpreting ranks among the most demanding cognitive tasks a person can perform. The interpreter must listen, comprehend, reformulate, and speak almost simultaneously, often navigating unfamiliar accents or rapid-fire sequences of numbers that outpace any note-taking. Researchers call this the “tightrope hypothesis” , during most assignments, interpreters operate at the very limit of their mental capacity, and accuracy drops the moment that load increases. The physical toll is real, too. Time pressure elevates stress and heart rate, and quality suffers in sessions longer than 30 minutes, which is why interpreters rotate roughly every half hour.

That pressure has intensified as work shifted online. According to Nimdzi, the interpreting industry is worth $11.7 billion, and an increasing portion now happens remotely: via video, through inconsistent audio, in hybrid meetings that barely existed five years ago. The captioning tools interpreters had access to were never designed for them. Intercall was created to fill that void.

Designed for the Interpreter

Intercall is a real-time, human-in-the-loop captioning and translation platform built exclusively for professional interpreters. Human-in-the-loop means the human remains in control; the software provides support. As the interpreter works, Intercall displays everything spoken on screen as text, in both languages, the instant it is uttered.

Founder Bahodir Rajabov started from an insight most products miss. An interpreter doesn’t need a polished transcript after a meeting; they need the right word in the moment, fast enough not to disrupt their concentration. The name they missed. The figure that flew by. The technical term from an unfamiliar field.

Rajabov began programming at 14 in Bukhara, Uzbekistan, and later worked on IBM’s generative-AI team. He built Intercall to solve a problem the industry had overlooked, and he was the primary architect of its core: the native audio capture, the low-latency transcription pipeline, the terminology system, and the cross-platform desktop workflow. The generic captioning systems available to interpreters lagged three to five seconds behind the speaker. That works for subtitles. It’s a serious problem for a medical interpreter relaying symptoms in real time, or in a courtroom where half a second can alter the meaning of a statement.

As Rajabov puts it: “Nobody had built anything for interpreters. They are masters of their craft, working with borrowed tools. Captions made for viewers, translation apps made for travelers. They just needed something made for them.”

How It Works

The breakthrough came from leaving the browser. Intercall runs as native software on the interpreter’s own machine, installed like Word or Zoom rather than opened as a website. Written in C++, it interacts directly with the operating system and captures audio straight from whatever call is active , whether Zoom, Microsoft Teams, or Google Meet. No browser extension, no meeting bot, no screen-sharing.

It performs three functions. It transcribes the conversation in real time, fast enough that the text stays synchronized with the speaker rather than lagging behind. “It has to feel instant,” Rajabov says. “The moment the text falls behind the speaker, it is useless, and interpreters simply close it.” It captures the parts most easily lost and most costly to get wrong , proper nouns and specialist terms , and lets interpreters load up to 600 of their own terms before a session: cardiology vocabulary before a heart appointment, case names before a hearing. And in multilingual calls, it automatically switches between languages and dialects, across dozens of them. The doctor’s English and the patient’s Russian, captured side by side, in one conversation, exactly as interpreted calls actually happen.

For most interpreters, that single window replaces a patchwork of tools built for other purposes: meeting captions, a notepad, a translation tab open in the browser. The difference shows in the work itself. Fewer requests for repetition. Long, dense passages relayed without breaking the conversation’s flow. Far less of the frantic note-taking that drains interpreters by the end of a shift, so the energy once spent catching and holding details goes back into the interpreting.

Intercall is built to assist, not to take over. It does not interpret for the user. It surfaces what would otherwise slip past and hands it back, leaving the interpreter in charge of the part only a person can do: meaning and judgment. That choice placed Intercall early into a human-in-the-loop category that analysts now track , AI that supports professional interpreters during live work rather than replacing them. Larger players have since followed the same path. Boostlingo added an AI productivity tool in 2024, KUDO brought assistance into its interpreter console in 2025, and analysts at Slator and Nimdzi now cover the category.

Confidentiality

Interpreters raise the obvious objection almost immediately: if it’s AI, is it storing my audio and training on my clients’ private conversations? The company’s answer lies in how the platform is built, not in a privacy policy. According to Intercall, audio, transcripts, and translations exist only in memory and vanish when the session ends; nothing is written to a server, and nothing is used to train the models. Whatever an interpreter generates stays theirs. It is encrypted and built for the confidentiality that HIPAA-regulated medical work and the courts demand , settings where a leaked transcript is a breach, not an inconvenience.

Adoption and Impact

Intercall has been live for over a year and a half. More than 3,000 professional interpreters across 18 countries have used it, and thousands of hours of live interpretation now run through the platform every month. What stands out is the reach. The largest user group is in the Dominican Republic, with paying interpreters also across the United States, Poland, Peru, Honduras, Mexico, Canada, and beyond, many working through interpreting providers including Propio Language Services, iCall International, and others. They use it on live assignments in hospitals, courtrooms, and conference halls. Its growth has come mostly through the profession itself: interpreters recommending it to colleagues and saying so publicly.

The impact they describe boils down to two words: efficiency and accuracy.

Efficiency, because the tool absorbs the part of the job that exhausts interpreters , holding names, numbers, and details in memory while still listening and speaking. Interpreters report working longer hours with far less fatigue. Some describe ending shifts with energy they used to burn on frantic note-taking. A few say it pulled them back from burnout that had nearly pushed them out of the profession. In a field where the working limit is mental fatigue rather than skill, that is the difference between a sustainable career and an early exit.

Accuracy is the other half. Everything said on a call is transcribed and translated in real time, in both languages, in front of the interpreter, so requests for repetition drop sharply and the details most expensive to get wrong , drug names, dosages, case numbers, addresses , stop slipping through. In a hospital, that means the patient who speaks no English receives discharge instructions exactly as the doctor gave them. In a courtroom, it means testimony enters the record complete.

One interpreter at Propio Language Services in Colombia, remote for more than five years, described working 48-hour weeks with paper notes and a Boogie Board: “By the end of the day, I often felt mentally exhausted from constantly relying on memory and note-taking.” Even with generic caption tools running, “I still had to manually write down phone numbers, addresses, names, and other critical details… the cognitive load was always there.” Then he found Intercall: “It has significantly reduced my mental workload and allowed me to focus on what matters most: interpreting accurately and serving my clients effectively. No more juggling a notepad, translators, and multiple reference tools. I genuinely cannot imagine working without it.”

Independent practitioners say much the same. In a detailed review, interpreter Nourhane Atmani wrote that it “does reduce unnecessary cognitive load, helps us catch what we might otherwise miss (names, numbers, etc.), and supports us when working in demanding or unfamiliar contexts,” while making clear that it “does NOT replace interpreters (and never aims to).” Another put it more plainly: “It is like working with a co-pilot. I am still flying the plane. Now I have better instruments.”

What Comes Next

Intercall is built and run by a lean, dedicated team, and the company is now focused on scaling: reaching more interpreters and bringing the platform to interpreting agencies and enterprise teams. The engineering work follows the same line, with most effort going into one stubborn problem: maintaining accuracy in the noisy, low-quality audio interpreters so often have to work with.

That discipline has earned the field’s trust. “We started as an assistant. We proved our accuracy and earned that trust first,” Rajabov says. “You cannot skip that stage. It is where the others went wrong.”

The interpreter stays at the center of the work , listening, weighing, and turning one language into another in real time. Intercall is not there to take that over. It is there, in Rajabov’s phrase, to give them a second pair of ears.