Why We Built OllaSync — A Better Zoom for the Learning Industry
Twenty years of running a classroom and a live online class at the same time, five problems that landed on the trainer after every single session, and the three months it took to build live translated speech from scratch.
About the Networkers Home Engineering Team
Our content is written by industry practitioners with hands-on experience in enterprise environments. We don't write theory — we share what actually works in production.
A Class Is Not a Meeting
Networkers Home has been in the training business for twenty years, and the shape of a working day has not changed much in that time. There is a classroom in Bangalore with students in it. On the trainer's machine, a Zoom session is running at the same time — for the two or three who could not travel in that morning, and for the much larger group who enrolled online and were never going to be in the room at all.
One class. Taught to a room and to a browser, simultaneously, by one person.
We used Zoom for this because there was nothing else. Not nothing better — nothing else. A training company needs a stable screen share for two hours, a recording that survives a dropped connection, and forty people joining from nine time zones on whatever laptop they own. For most of the last decade Zoom was the only product on earth that did all of that without falling over, and it deserves enormous credit for it. A great deal of what the online education industry became after 2020 exists because Zoom worked.
It just was not built for teaching. It was built for meetings. And the difference between those two things is the entire reason OllaSync exists.
The whole argument in one line
When a project call finishes, it has already achieved what it was for. When a class finishes, somebody still has to record who attended, write up what was covered, get those notes to thirty students, and work out whether the four people who never turned their camera on understood any of it. In a meeting-shaped tool, all of that lands on one person: the trainer, in the evening, by hand.
The Five Things That Kept Landing on the Trainer
Five jobs, after every class, every day, for twenty years — none of which is teaching.
Attendance. The wrap-up. The notes. Getting the notes out. And the language gap.
None of these are exotic problems. That is exactly what made them maddening. Every one of them is the kind of thing software has solved in a dozen other industries, and every one of them was still being done by a tired person at nine in the evening.
Attendance. Not for bureaucracy — for teaching. If a student has missed the last three sessions on OSPF, that is something the trainer needs to know before the fourth one starts, not at the end of the month. Zoom can give you a participant report. It lives in an admin portal, on the right licence tier, and somebody has to go and get it.
The wrap-up. A training operation running several batches a day needs a short answer to "what happened in the 9am class?" without phoning the trainer. That answer did not exist anywhere except in the trainer's head.
The notes. Every serious class produces them. Almost no class produces them consistently, because writing them up is the last thing anybody wants to do after teaching for three hours.
Getting the notes out. This was the stupidest one. Notes would get written and then sit in a document, because sending them meant copying thirty email addresses out of an enrolment sheet.
The language gap. The one that actually mattered, and the one we spent the most on.
The Language Gap Was the Real One
Networkers Home teaches students across India, the Gulf, North and West Africa and Latin America. A very large number of them read English at a professional level — they can work through Cisco documentation without help — and follow spoken English badly. Those are not the same skill, and the gap between them is where students quietly fall behind.
It is worse than general spoken English, too. This is spoken English at teaching pace, in a Bangalore accent, about BGP route reflectors, with the trainer talking over a terminal window. A student in Casablanca joining at two in the morning is not struggling with the protocol. He is struggling with the audio.
Captions help. We ran captions for a long time and they are genuinely useful. But ask anyone who has relied on them what it is actually like: you are reading text at the bottom of the screen while a live lab demonstration happens above it, and you have to choose which one to look at. Choose wrong and you miss the command. Captions are a real improvement. They are not the same as hearing it in your own language.
The thing nobody says out loud
Why This Had to Be AI, and Not Process
The realisation did not come from a classroom. It came from a different kind of call entirely.
Through last year Networkers Home was building complex AI products for clients in the United States. Long engineering calls, most of them on Zoom, most of them at difficult hours. And somewhere in the middle of those months the contrast became impossible to ignore. We were shipping systems that could transcribe a conversation, summarise it, translate it and route the follow-up work automatically — for a customer in Texas. Then we would close the laptop, walk back to our own training operation, and find a trainer typing up class notes by hand and emailing them to thirty students one address at a time.
We were solving this problem professionally, for other people, and living with it ourselves.
The process answer would have been to hire people. A coordinator per batch to take attendance and chase notes. Interpreters for the languages with enough students to justify one. We costed it. It does not work — not because of the money, but because of the shape of the problem. You cannot hire your way out of sixteen languages across a dozen time zones for classes that run every day. The moment a language has only four students in it, a human interpreter is impossible and those four students are on their own.
AI does not have that shape. The cost of supporting the sixteenth language is roughly the cost of supporting the second.
Building Live Translated Speech From Scratch
Translated captions are a solved problem. You can wire them up in an afternoon. Live translated speech — where the student hears the class in their own language while the trainer is still talking — is a different category of engineering, and the naive version of it is unusable. It took more than three months to get from working to tolerable, and almost all of that time went on problems that do not appear in any demo.
Three months of gotchas
The latency budget
Past a certain delay it stops being a class and becomes a broadcast, because the student can no longer interrupt. Every component — capture, recognition, translation, speech — had to fit inside a budget small enough that a raised hand still works.
Turn-taking
Knowing when a sentence has actually ended, as opposed to the trainer pausing to type a command. Get this wrong and the translated voice talks over the trainer, or waits so long the moment has passed.
When to commit a partial transcript
Speak too early and you translate a half-sentence that changes meaning when it finishes. Wait for certainty and you blow the latency budget. This single trade-off consumed more time than anything else.
Locking the technical vocabulary in English
A CLI command is not a phrase to be translated, it is a string to be typed. So are vendor feature names. Translating them produces confident, fluent, completely wrong instructions — and a student who types one into a real switch gets an error and loses faith in the tool.
Making it survive a real classroom
Trainers interrupt themselves, answer questions mid-sentence, and read commands aloud while typing them. A model that only works on clean, turn-based speech does not work on teaching.
Where this puts us
The result runs today across sixteen languages, including Spanish, French, Chinese, Japanese, Hindi and English. You can see live translated speech running on real sessions rather than a recorded demonstration.
The Cost Problem, and Why It Decided Everything
A feature that works but costs more than the course fee is not a feature. Early on, running this through a general-purpose realtime API made the economics absurd — a three-hour class with a dozen language streams cost more in inference than several students had paid to attend it. At that price you ration the feature, and a rationed feature may as well not exist.
The pipeline was rebuilt for one job instead of every job. A trainer speaking to a class is a narrow, predictable workload: one dominant speaker, a known technical vocabulary, long stretches of continuous speech. A general realtime API has to be priced for every possible conversation. Ours only has to handle this one.
Measured against OpenAI's Realtime API on the same teaching workload, that brought the cost to roughly a twentieth. That is our own measurement rather than a published benchmark, and it is the number that turned live translation from a premium add-on into something we can leave switched on for every class, in every language, by default.
What OllaSync Does After a Class Ends
The translation is the hard engineering. The five things below are what actually changed the working day.
Attendance, per session. Every finished class shows who came — names as they appeared in the meeting, the host with their email — along with the start time, the duration and the count. It is on the meeting row's own menu, not buried in an admin console, because the person who needs it is the trainer.
A wrap-up email to the host, sixty seconds after the session ends. Attendees, minutes, recording status and a dashboard link. Nobody requests it and nobody exports it — it is simply there when the trainer closes the laptop. This is on by default for workspaces created from 11 September; older workspaces switch it on.
Automatic notes. When a recording completes, the transcript and the structured notes are written with no clicks at all. Not a button that generates notes — no step. Any manual transcription flows into the same notes while the switch is on.
Sending those notes to students. One button on the Notes panel emails them to everyone the host invited by email, plus any addresses typed in at the time. The invite roster is kept for exactly this purpose. What goes out is structured rather than a wall of transcript: a TL;DR, topics, concepts, homework, action items, decisions and the Q&A.
The invite note. A personal message field on the email invite. It is the smallest feature on this list and it changes attendance more than it has any right to. An invite that says something specific about today's session reads like a person. A bare calendar invite reads like spam.
Three different behaviours — worth being precise
Zoom and OllaSync, Side by Side
This table needs one honest caveat, because the comparison is easy to overstate. Zoom can do most of the things in the left column. It has participant reports, it has AI Companion summaries, it has a paid translation add-on. The difference is not that these capabilities are missing — it is that they are scattered across an admin portal and several paid tiers, designed for meetings, and every one of them needs a human to go and retrieve it after class. The comparison is about the workflow, not the feature list.
The comparison is about who does the work after class — not about which product has more features.
| After the class ends | Zoom | OllaSync |
|---|---|---|
| Who attended, and for how long | Participant report in the admin portal, on the right licence | On the session's own row — names, host email, start, duration, count |
| A summary of what happened | AI Companion summary, where enabled on the plan | Emailed to the host 60 seconds after the session ends |
| Transcript and structured notes | Transcript available; summaries on paid tiers | Written automatically when the recording completes — zero clicks |
| Getting the notes to students | The trainer copies, pastes and emails them | One button — goes to the invite roster plus typed addresses |
| A student who cannot follow spoken English | Translated captions, paid add-on, limited languages | Live translated speech and captions, 16 languages |
| A personal line on the invitation | Not part of the standard invite flow | Invite note field on the email invite |
What This Opens Up for Every Other Training Company
Networkers Home built this for itself, and it would have been worth it even if it had stopped there. But the five problems in this article are not Networkers Home problems. They are learning-industry problems. Any institute running live classes has a trainer doing admin in the evening, and almost all of them have students who are quietly failing on audio rather than on the subject.
The part that should interest other training companies most is the language one, because it changes what your market is. An institute in Pune teaching in English sells to people who are comfortable in English. The same institute, with the same trainers and the same syllabus, running live translated speech, can sell to francophone West Africa and to Latin America without hiring a single new instructor. That is not a feature upgrade. That is a different addressable market, reached with the staff you already employ.
That is why OllaSync is a product rather than an internal tool.
Why One Group Builds All of This
A training institute does not normally ship a realtime speech infrastructure product. The reason this one could is that the engineering was already in the building.
Networkers Home is the oldest company in the group and the one that generates the problems, because it runs real classes for real students every day. OllaSoftware is the parent company that builds the answers. OllaSync is the one this article is about. Around them sit OllaNode, OllaSuper, OllaCore and OllaGraph, each solving a different part of the same infrastructure problem.
The useful consequence is a short loop. A trainer hits a problem in a Tuesday class, and the people who can fix it are not a vendor's support queue — they are in the same organisation, and the class next Tuesday is the test environment. Very little software for the training industry is built by people who have to teach on it the following week.
What OllaSync Is Not
An article like this is worth very little without the other half, so here is the other half.
Where OllaSync is the wrong tool
- ✕The translation is machine translation. It is very good. It is not a human interpreter, and it will occasionally produce a phrasing a native speaker would not use.
- ✕Command syntax and vendor terminology stay in English on purpose. If you expected a fully localised experience down to the CLI, this is not that, and it never will be — translating a command makes it wrong.
- ✕It is built for teaching, so it is shaped around a dominant speaker and a class. For a 400-person all-hands or a free-for-all brainstorm, a meeting product will serve you better.
- ✕Notes are generated from what was said. A session where the trainer works in near silence at a whiteboard produces thin notes, because there is little to work from.
- ✕It will not fix a class that was badly taught. Perfect attendance records of a poor session are still a record of a poor session.
Networkers Home still runs classes where the right answer is a room, a whiteboard and a rack of real equipment. That has not changed in twenty years and it is not going to. What changed is everything that used to happen after the class ended.
If you would like to see what this looks like from a student's side, our one-on-one online training runs entirely on OllaSync — and the trainers teaching those sessions are the same people who reported the problems this article describes.
Questions We Get Asked
How is this different from Zoom’s translation add-on?
Zoom’s add-on produces translated captions — text on screen, in a limited set of languages, on a paid tier. OllaSync produces live translated speech as well as captions, so a student can listen in their own language rather than reading while also trying to follow a lab demonstration. Reading captions and watching a terminal at the same time is the specific problem we set out to remove.
Does the trainer have to speak slowly or differently?
No, and that was a hard requirement from the start. A trainer who has to change how they teach will stop using the tool by the second week. Trainers at Networkers Home teach exactly as they always have, including interrupting themselves and answering questions mid-sentence.
What happens to command syntax and vendor terminology?
It stays in English, deliberately. Translating a CLI command or a vendor feature name makes it wrong, and a student who types a translated command into a real switch gets an error. OllaSync holds that vocabulary in English and translates the explanation around it.
Can other training companies use OllaSync?
Yes. OllaSync is a product, not an internal tool. Any institute running live classes — technical or otherwise — can use the same attendance, notes, wrap-up and translation workflow. Details are at ollasync.com.
How is it a twentieth of the cost?
That figure is our own measurement of the same teaching workload against OpenAI’s Realtime API, not a published benchmark. The saving comes from building the pipeline for one job — a trainer speaking to a class — instead of using a general-purpose realtime API priced for every possible job.
Does anyone need to install anything?
No. Students join from a browser in one click. This matters more than it sounds: a large share of our students join from shared machines, from office laptops they cannot install software on, or from phones on mobile data.
