We provide compute and run your training job on it. You supply the code, the model design and the dataset. We provision the hardware, resolve environment and dependency problems, run the job to completion, watch it while it runs, and hand back every artifact it produces.
We don't design your model, write your training code, clean or label your dataset, or write your analysis. That part stays yours. Universities differ on what they allow for external compute, so it's worth checking your department's rules before you submit.
We guarantee your job executes to completion on working hardware and that you receive the artifacts. We don't guarantee accuracy, loss, mAP, F1, perplexity or convergence — those depend on your data and architecture, which we don't modify.
A run that completes with disappointing metrics is a delivered job. It isn't refundable on that basis.
Every job starts with a Rs 750 sanity check — one epoch, to confirm your dependencies resolve and your code reaches the training loop, and to measure the real per-epoch time your quote is built from.
Your quote is measured epoch time × your epoch count, plus a 30% buffer.
If a run looks like it will exceed that buffer, we stop, save the latest checkpoint and ask you. You choose: pay the difference and continue, or take what's been produced. We never run past the buffer and bill you afterwards. If you change epochs, dataset or config after the quote, we re-quote.
Staged, so you never pay a large amount before seeing real work.
Accepted: bank transfer (Commercial Bank, Sampath, BOC, HNB), FriMi, Koko. Koko splits any amount into three and adds 4% for their fee. Bank transfer has no fee and confirms fastest. We don't do informal instalments — the stages above are there so you don't need one.
Student rates are roughly half the research and commercial rate, and cover undergraduate and taught masters coursework including final year projects. They don't cover funded research, publication-bound work or commercial jobs. We don't ask for proof — tell us which applies and we take your word for it. Booked 14+ days ahead: 15% off.
Referrals. If someone you referred completes a paid job and names you when they submit, you get Rs 1,500 off your next run. Credit applies to one future job, doesn't stack with itself across multiple referrals on the same job, and expires 6 months after it's earned. It can't be paid out as cash.
Full refund if: hardware failure or a provider outage stops the run and we can't restart it; we can't get suitable hardware in the time agreed; or we decline your job after you've paid.
No refund if: the run completed but the metrics disappoint; your code, data or config caused the failure; or you cancel after the full run has started.
Jobs run on third-party cloud infrastructure, currently RunPod and Vast.ai. Your code and dataset are uploaded there for the duration of the run. Some of that is community hardware run by independent hosts.
If your data is unpublished, confidential, personally identifying or under ethics approval, say so in the form. Those jobs run only on isolated secure-cloud instances, which cost more and are quoted accordingly. If you don't tell us, we'll assume standard handling is fine.
Anything illegal under Sri Lankan law, cryptocurrency mining, or password and key cracking. Self-service instances are monitored for these. If we find them the instance stops, remaining hours are forfeited, and we won't take further jobs from you.
Our liability for any job is capped at what you paid for it. We're not liable for missed deadlines, lost marks, lost data or the outages of third-party providers.
We're a small operation. We'll tell you early and honestly when something goes wrong, but we can't underwrite the outcome of your project.
We may update these terms; the version that applies to your job is the one published when you submitted it. Questions before you submit — just ask. Much cheaper for both of us than a disagreement later.
WhatsApp +94 75 924 0520 · Email nevaracloud@gmail.com
NevaraCloud