Four questions, answered in order
The page to read if you want the whole thing in one sitting, in commercial language rather than engineering language. Every section ends at the page that proves it, and no claim here is stronger than the claim on that page.
1 — What we have built
A complete B2B streaming platform, built to be handed to an operator rather than rented to one. The catalogue and the admin console the business runs on. Live channels and on-demand, on separate delivery paths because they fail differently. Encoding into an adaptive ladder. Subscriptions, coupons, trials and one-off purchases, priced in taka. Applications for web, Android phone and Android TV, published under your own developer account. And playback telemetry from every session, so quality is a number rather than a complaint.
It runs a live consumer service today, which you can open in a browser without asking us. What it does not do is also published, on the same page and in the same table: 60 capabilities with a status on every row, of which 23 have no code at all. A platform that publishes its gaps is telling you what the first three months of an engagement will actually be about.
The three shapes it is bought in are a licence, an implementation engagement, or a managed agreement. The platform underneath is identical in all three and you can move between them without changing platforms.
2 — How the cost is kept down, and kept down
Streaming has three meters: turning a master into renditions, moving bytes to a screen, and knowing what the screen did. Rented, all three bill by usage. That is comfortable at launch and it is the reason a growing service can watch its margin fall while its audience rises — every new viewer arrives as a line on next month's invoice.
Owned capacity replaces the meter with a commitment you size once. Inside it, the next gigabyte costs nothing. The bill steps when you renegotiate the commitment rather than when your audience grows, which is the shape a subscription business needs and the opposite of what it usually gets.
The second half of that sentence — *and keep it down* — is the part worth checking rather than believing. On the South Asian rate card, tripling delivered volume multiplies the rented bill 2.89× and improves the blended rate by 3.8%. Scale is not a lever a regional operator can pull, because the volume discount that makes rented delivery work elsewhere barely exists here. Those are the vendor's own published numbers, dated, with the arithmetic shown.
What owning costs is on the same page and belongs in the same breath: capacity is bought before it is needed, utilisation below plan is money already spent, and hardware fails. Below roughly a terabyte a month the meter is genuinely cheaper, and the calculator on the pricing page is allowed to tell you so.
3 — How the service is guaranteed
Support is 24/7, and that is a commitment rather than an aspiration. When something breaks you reach an on-call engineer directly rather than a queue that opens in the morning; if a critical incident passes its update interval without a workaround, the engagement lead is brought in; and the last step is the owner of the business, named in your agreement. Three steps, because there are three — a four-tier ladder would look more impressive and would be fiction.
Severity is assigned from what your viewers experience. Service down or playback failing for most viewers is critical. Payments failing at checkout, or one live channel dark, is major. That definition survives a bad night in a way that 'how urgent did the email sound' does not.
Response, update and resolution targets are published in full rather than withheld until a sales call, and every one of them is marked indicative until it is written into your agreement. What is deliberately not published anywhere on this site is an availability percentage. Nothing measures availability continuously today, so any figure would be an assertion; and a percentage is a contractual liability with a credit schedule behind it, which the owner signs rather than a website states.
There is also a structural half to this answer that no support agreement provides. The estate runs on your infrastructure with the Terraform and the runbooks in your hands, so the platform does not depend on the continued availability of any particular engineer, ours included.
4 — How the hybrid model actually works
Split the platform by which half is expensive to rent. The application tier — the APIs, the catalogue service, the console, the sign-in and billing paths — is elastic, spiky and cheap to run on somebody else's compute. It goes on AWS, where absorbing a traffic spike is what you are paying for and the bill is small.
The other half is the three meters. Encoding bills per output minute per rendition, with no volume discount at all. Origin storage bills per gigabyte-month against a catalogue that only grows. Delivery bills per gigabyte against an audience you are trying to grow. Those stay on hardware you own, where a GPU that has been bought costs less per minute every time it is used, a cache miss costs electricity rather than a line item, and the marginal gigabyte inside a transit commitment is free.
One number makes the encoding half concrete. A full pass over a 1,000-hour catalogue bills $3,060 on rented transcoding, and it bills again in full every time you change the ladder or add a codec. On a GPU you own, the same pass is a scheduling decision.
What makes the split practical rather than theoretical is that nothing is tied to either side. Origin storage is S3-compatible, every service is containerised, and the whole estate is declared in Terraform — so the same platform runs fully on-premise, fully on AWS, or split, and moving between them is a change of target rather than a rewrite.
That is the whole argument
If it fits what you are trying to build, the next step is your numbers and an architecture back. If it does not, the spec sheet will have told you why faster than a call would.