01about us

We built this because we needed it ourselves

Vidora Labs was not a market we spotted from the outside. It was a ticket in our own backlog that nobody could close at a price that made any sense.

02where this came from

A ticket we could not close

We build and operate streaming platforms — apps for Android, Android TV, Fire TV and iOS, the backends behind them, the EPG pipelines, device pairing, payment reconciliation. The unglamorous machinery that keeps a channel on air.

On one of those platforms the audience turned out to be far more multilingual than the content. People were watching in a language they only partly followed, and every month the same request came back: subtitles. We priced it with three vendors. The quotes came back in the tens of thousands and the schedules came back in months. For a regional operator that is not a procurement decision — it is a no.

So we did what engineers do and wired a speech API to a translation model over a weekend. It produced a file. It did not produce subtitles. Cues ran four lines long and sat on screen for eleven seconds. Hebrew rendered backwards the moment an English brand name appeared. Every second line addressed a woman in masculine grammar. That gap — between a file and a usable subtitle track — turned out to be most of a year's work. This is that work, made available to everyone with the same ticket open.

03how we work

Six decisions we made early and have not reversed

Explain the mechanism

The engine page walks through every stage, the numbers each one enforces and where it falls over. You should be able to judge whether this will work for your catalogue before you spend anything.

Flag, do not hide

Every cue carries a confidence score and weak ones are surfaced. A vendor returning uniformly confident output on a difficult film is not more accurate — just less honest.

Name the limits

There is a section on this site listing what we are bad at. It costs us some deals and saves everyone the discovery call where it comes out anyway.

One price, published

No tiers, no "contact sales for pricing", no enterprise page that hides the number. You can work out your bill before you make an account.

Hard scripts are not a roadmap item

Hebrew, Arabic, Punjabi and Tamil were in the first release. We test right-to-left output in real players, because that is the only place the bugs show up.

Operators, not a demo

We run production systems and get paged by them. That shapes what we build: retries, resumable work, honest failure states rather than a polished happy path.

04the company

Rank First Technologies

Vidora Labs is a product of Rank First Technologies, a software company headquartered in Mohali, Punjab, with offices in Beverly Hills and Surrey, British Columbia. We build enterprise software across streaming, CRM, HR systems, ERP and AI voice — for our own products and for clients who need something built properly rather than assembled from templates.

That range matters more than it sounds. A subtitling service is not really a machine learning product; it is a distributed job system with media handling, a queue, retries, storage lifecycle rules, per-account billing and a ledger that has to reconcile. Those are the parts we have been shipping for years. The models are the easy bit.

OTT and IPTV platforms Android and Android TV iOS and tvOS Django PHP Kotlin AWS
05track record

Things we run in production

Not case studies with invented percentages. Systems that are on right now and page us when they are not.

A multi-platform OTT service

Android, Android TV, Fire TV and iOS clients against a shared backend. Live channels, VOD, EPG ingest, device session management, TV code pairing, casting, and the release pipeline that ships all of it.

A college ERP with live finance

Admissions, examinations, hostel finance, instalment plans and payment gateway reconciliation — running against real money and real deadlines with a small team on call for it.

A multi-entity HR platform

Asset management, approvals, a workflow engine, geofenced attendance, document generation and analytics across several legal entities, with the hardening a system holding employee data needs.

Voice-driven sales systems

CRM with automated outbound calling, live transcript monitoring and generated estimates — which is where a lot of our speech-pipeline instincts came from before this product existed.

06quality

What we actually measure

Word error rate is the metric everyone quotes and it is close to useless on its own, because a transcript can be word-perfect and still produce an unpublishable subtitle track. We track four numbers per job and you see all of them.

Recognition confidence
Per word, aggregated per cue. Tells you where the audio beat the recogniser.
Conformance violations
Cues that breached the reading-speed, duration or line-length rules before correction — and how hard the correction had to work.
Condensation ratio
How much text had to be cut to fit the available time. Heavy condensation is where meaning quietly disappears.
Flag rate
The share of cues sent to review. Typically 10–15% on features, 3–6% on interview and lecture content.
07your media

Your masters are not training data

Deleted within minutes

The media file is removed as soon as the transcript comes back — usually minutes after upload, not days. Only text persists after that.

No training on your content

We do not train models on customer media, and we run every processing step in zero-retention mode where that option exists. Your unreleased title does not become anyone's dataset.

Processed, then discarded

Audio exists in our systems only as long as it takes to produce a transcript. Nothing is archived, nothing is warehoused, and there is no second copy sitting somewhere for later.

Your files, your call

Finished subtitle files stay in your account so you can re-download and re-export without paying again. Delete them whenever you like and they are gone.

08where we are

Three offices, one on-call rota

Mohali, India

Engineering · headquarters

Where the pipeline is built and where most of the team sits. Support hours start here each day.

Beverly Hills, USA

Distribution relationships

Closest to the customers who care most about deliverable specs and territory requirements.

Surrey, Canada

North American operations

Covers the overlap window, so an incident at three in the morning in Mohali still has someone awake on it.

09what is next

The roadmap, honestly labelled

Glossaries and character sheets
Feed character names and their grammatical gender into the translation prompt. The single biggest quality gain available to us. Building now.
Series memory
Character sheets that persist across a whole series automatically, so episode nine knows what episode one established.
A public API
The application already runs on one internally. Exposing it is more a documentation problem than an engineering one.
Custom conformance profiles
Matching a distributor's spec sheet rather than our defaults. Available on request today, not yet self-serve.
Lyric handling
Currently we transcribe songs and leave them. Doing it properly is a craft problem, not a compute problem, and we do not have a good answer yet.
Lip-synced dubbing
People ask. We are not building it and we are not pretending it is coming next quarter.
10careers

Who we are looking for

Small team, no layers, everyone touches production. If you want to be handed a spec and left alone, this is the wrong place. If you want to own a system end to end, it is a good one.

  • Backend engineers who have run queues at volume and have opinions about idempotency
  • Frontend engineers who care whether a subtitle renders correctly in six scripts
  • Native reviewers for Hebrew, Arabic, Tamil and Punjabi — part-time, remote
  • A support engineer for the India-morning window who can read a stack trace

No listings page yet. Write to support@vidoralabs.cc with something you have built and we will read it properly.

11getting started

What the first week looks like

There is no onboarding call and no implementation fee. You will be running a real title within an hour of signing up.

  1. Run one difficult title

    Not your cleanest interview footage — the one with the score under the dialogue and three people talking over each other. That is the one that tells you whether this works for your catalogue.

  2. Put it in front of a native speaker

    The only evaluation that means anything. Do it before you commit budget, not after.

  3. Check the flag rate

    If it is high, that is the pipeline being honest about difficult audio, not hiding it. Decide whether reviewing that ten percent is worth your time.

  4. Batch the backfill

    Load the wallet with whatever you want to spend this month and work through the catalogue at that pace. The minimum is $100.00 and nothing expires if you slow down.

12talk to us

Awkward requirement? Say so.

Custom reading-speed profiles, a source language we have not listed, a distributor spec sheet, volume that makes the flat rate silly — all of that is a conversation, not a support ticket.

Ready in minutes

Run a real film through it tonight

Create an account, top up the wallet, and process a title you actually care about. A feature film in two languages costs about $16.00, so you can judge the output on real material rather than on a demo clip.