AI features, tested before the balance is due.
Support desks, WhatsApp and voice assistants, automations, websites and full products. Choose the scope, see the price, and send a brief. Each build is scored on your own questions before the second half is due.
- 150question test, re-run on every content change
- 41 / 41unrelated questions declined in the latest demo run
- 14working days from kickoff to live
- 0platform fees. The code is yours.
A Scale sprint, day by day. Half is paid at kickoff; the balance is due once the day 9 test is passed.
What you receive.
Five deliverables ship with every build. The views below come from test runs against public help pages, with company names masked.
The answer desk, on real questions.
Choose a question from a test run to see its recorded answer and the page it came from, or type your own. If the pages do not cover it, the desk passes it to a person.
Six templates to start from.
These previews are quick, in-page mini-apps with sample data. The real versions run on a server, database and language model. Start a project from any template and I replace the sample data with yours, or ask for a free demo on your own site.
Two sprints, one fixed price each.
Prices are in rupees, before your own AI and messaging usage fees. No GST is charged. Open an item to see what it includes.
Outside India?
Smaller, entry-scope versions of some of this work are sold on Fiverr and Upwork, paid in dollars with the marketplace's protection. The full builds above are quoted direct.
- AI chatbot trained on your docsfrom $49
- WhatsApp AI assistantfrom $59
- Full-stack AI SaaS appcustom quote
- Fiverralokkashyap824
What an answer desk saves.
Enter your own figures. The shaded band on the deflection slider marks the range the test runs reached.
This is arithmetic on your inputs, not a promise. "Answered by itself" means answered with a source in a test run; how often those answers are right is measured on your own questions before handover. Model usage is billed to your account and is a few paise a question.
Five test runs, every score published.
Each run read one company's public pages and answered 10 to 12 questions, two of them deliberately off-topic. No setup, no tuning. Select a run to see its scores.
Scope a project.
Uses the same prices and discount rules as the cart. Nothing is sent or charged.
Try it on your site.
Give me the pages to read and what your customers ask most. The form drafts a message for you to copy or open in your mail app. Nothing is sent from here.
Included in every build.
Scored before handover
A question set drawn from your customers, marked answer by answer. Pass marks are agreed in advance and you receive the full report.
Answers from your content
If your content does not cover a question, the desk passes it to your team.
You own it
Code in your repository, a runbook and a team walkthrough. No platform fee.
Privacy
Personal details are masked before anything reaches an AI model. Data handling terms under the DPDP Act, in writing.
Fourteen working days, in order.
A Launch sprint runs the same way in 10. Half is paid at kickoff; the balance is due once the day 9 test is passed.
working days
I'm Alok. You work with me directly.
I'm an AI engineer in Punjab, finishing an MCA in AI and machine learning at LPU. I built Vachix, an AI voice companion app, from end to end: more than 110 API endpoints, backed by over 700 automated tests.
Everything here is built on Foundry, the delivery toolkit I made and test on every change, with more than 1,300 tests. Your project starts from working parts on day 1, not a blank page.
Questions before you start.
Do I pay when I check out?
No. Checkout creates a project brief with your prices. We confirm scope on a short call first; you pay half to start and the rest after it passes its test. MVPs are billed 40% at start, 30% at a mid-build demo and 30% on acceptance.
What if it gives a customer a wrong answer?
It only answers when it can point to a line in your own content; otherwise it hands the question to your team. Misses on the test set are fixed before launch. Wrong answers you report later are fixed within 3 working days during the support period.
What do the monthly care plans cover?
Monitoring, content refresh, a monthly re-run of the test and cost checks. They are not a bug-fixing hotline: the Watch plan is capped at about 4 hours a month. Model, hosting and messaging costs pass through at cost on your own accounts.
What does AI usage cost on top?
Model and WhatsApp usage is billed to your own accounts by those providers. For a support desk that is usually a few paise per question at current prices. I estimate it before we start.
Is the desk demo above a live AI model?
No. The suggested questions replay answers recorded in real test runs. Typed questions are matched by keyword against a small offline sample of captured passages and quoted back. A real build uses a language model on your own pages.
Who owns the code?
You do, once it is paid for. It lives in your repository, with no platform fee and no lock-in.
Can you do something that is not listed?
Probably. Add the closest item, describe what you need in the notes at checkout, and I will quote it.