Yes, up to 45 days after the invoice, but only on the Premium SLA. The refund is prorated to unused service days and the request must be made in writing to billing.
Demo archive · fictional company · hover a source
Document AI & RAG engineering · Founder-led studio · Est. 2026
AI thatshows itssources.
We build assistants that answer from your documents and prove every answer with the exact page it came from. If the documents don't say it, neither will they.
Ask the archive
Can an enterprise client get a refund after 30 days?
Byte by byte, back to the source. That is what Zentrix means to us, and we build it into every answer.
The name, explained
Byte Zentrix
- 1byte The smallest piece of information. Every answer we build starts from one, and can be traced back to it.
- 2zen Calm and clarity. What a team feels when it stops digging through four hundred PDFs.
- 3-trix, from matrix The grid where everything connects: your documents, your data, your workflows.
Put together: calm, connected order in the middle of scattered data. We build AI that answers from your documents and traces every answer back to the exact document, page and line it came from. Byte by byte, back to the source.
Folio II · The problem
Most AI answers sound confident. Few can prove it. We build the ones that can, and the ones that say “I don’t know” when they can’t.
Hover or tap a card
Without a source → Traced to source
- Without proof · i
It invents a refund policy that sounds exactly right.
Traced · iIt answers only from the retrieved passages, or says the documents don’t cover it.
- Without proof · ii
It lists ten files as “sources” and hopes you won’t check.
Traced · iiEach claim cites the exact document, page and line, one click away.
- Without proof · iii
Accuracy is a feeling in a demo.
Traced · iiiAccuracy is a score, measured against your team’s real questions before handoff.
Interlude · Listen
Hear it answer, and hear it refuse.
A client asks two questions. One is in her documents, one isn't. Watch what the assistant does with each.
- Sara
We have four thousand pages of contracts and policies. My team spends hours hunting for one clause. Can your assistant actually help?
- Zentrix
Yes. Ask me anything about your documents. I'll answer, and show you the exact page I used.
- Sara
Okay. Can an enterprise client get a refund after thirty days?
- Zentrix
Yes, up to forty-five days, but only on the Premium SLA. That's in the Master Service Agreement, page twelve, section seven point two.
- Sara
And clients in Germany? Is there a separate policy for them?
- Zentrix
I couldn't find a Germany-specific policy in your documents, so I won't guess. The closest source is the general Refund Policy, page four.
- Sara
That's the part I like. It tells me when it doesn't know.
Illustrative conversation · voices generated live by your browser, nothing downloaded
Folio III · Selected work
Systems that cite, shipped.
Three builds, each with the numbers that mattered. No invented logos, no borrowed case studies.
Robotics AI Tutor
RAG over dense technical manuals
An AI tutor that answers robotics questions from technical documentation, streams the answer and points to the passage it used.
4.2s → 1.2sresponse time
Study Material Assistant
Client build · answers with citations
An AI assistant trained on a student's own study material that shows exactly which page every answer came from.
3 daysto a working version
HealthMate
Lab reports → plain-language summaries
A family health platform that reads uploaded lab reports and medical PDFs and explains them in plain language, tied back to the values in the report.
1 uploadto a readable summary
work
Folio IV · The ledger
Fixed scope. Fixed price.
Starting small on purpose.
Every engagement starts with a paid one-week sprint, so you see a working prototype on your own documents before committing to a build.
- IDiscovery Sprint1 weekfrom $300credited to the buildRequirements, architecture and a working prototype in one week.¶ Requirements document written in plain English¶ Architecture diagram and stack recommendation¶ Working prototype on a sample of your documents¶ Accuracy notes: what the prototype got right and wrong
- IIDocument AI Assistant2–3 weeksfrom $1,500typical range $1,500–4,000A private assistant that answers from your documents and cites every source.¶ Ingestion pipeline for PDFs, Word, Google Docs, Notion or web pages¶ Hybrid retrieval (vector + keyword) tuned on your own questions¶ Answers with page-level citations and a source preview¶ "Not in the documents" refusal instead of guessing
- IIIAI Agent & Automation3–4 weeksfrom $2,500typical range $2,500–6,000Multi-step agents and n8n workflows wired into the tools you already use.¶ Agent design: tools, permissions, and where a human approves¶ Integrations with Gmail, Slack, Sheets, CRMs, databases or your API¶ n8n workflows or a custom Python/Node service¶ Step-by-step logs and a run history you can audit
- IVCare PlanMonthlyfrom $300/motypical range $300–1,000/moMonitoring, updates, new documents and accuracy checks every month.¶ Uptime and error monitoring¶ Monthly accuracy check against your evaluation set¶ Adding new documents and sources¶ Model and dependency updates
Market reference, 2026: US freelance RAG and agent specialists bill $130–300/hr. You get a fixed quote instead.
Folio V · The chain of work
Every step leaves a record.
The same principle as the software: nothing you have to take on trust. Every milestone produces something you can open, run or check.
Discovery Sprint
A call, a sample of your documents, and a working prototype by Friday. You get an accuracy report and a fixed quote.
↳ You decide with evidence, not a pitch deck.
Build, in the open
A demo every week on your real documents. Milestone payments only after you approve what you've seen.
↳ Weekly Loom + live demo.
Measure
An evaluation set of your team's real questions, each with its correct answer and source. The system is scored against it.
↳ Accuracy is a number, not a feeling.
Handoff & care
Repository, deployment access, documentation and a recorded walkthrough. Then an optional monthly Care Plan.
↳ You own everything. No lock-in.
What a client said
Delivered a working version in 3 days. He flagged risks early, explained technical trade-offs in plain English, and even told me when one of my ideas would break the accuracy. He was right. Exactly what you want in an AI developer.
Founder · 2026Fig. 1 · The whole team, honestly.
Folio VI · Who you work with
No account managers.
You work with me.
I'm Hammad Mustafa, an AI engineer in Karachi. I started Byte Zentrix to build AI systems that can show their work, after seeing too many chatbots answer confidently and wrongly.
You talk to the person who designs and writes your system. That means no 2–3× agency markup, no handoffs between people, and one person responsible for whether it works. Before the studio I built production dashboards and APIs at Iotfiy Solutions, and I currently ship full-stack work for a Dubai-based team.
- BasedKarachi · UTC+5
- OverlapUS mornings · UK afternoons
- StackPython · FastAPI · Node · Next.js
- FocusRAG · citations · agents