FAQ
Questions, answered.
The same questions come up on every discovery call. Here are the answers, written down so you don’t have to ask.
01
Engagement basics
How do projects start?
Every engagement starts with a 60-minute scoping call. By the end, we send a written technical brief — scope, architecture, risks, and a fixed-price proposal. No commitment until you sign.
Do you charge for the discovery call?
No. The first 60-minute call and the written brief that follows are free. If we don't think we're the right fit, we'll tell you and recommend who is.
How are projects priced?
We default to fixed-price scoped engagements with named milestones, not hourly. For longer engagements or ongoing iteration we offer monthly retainers with a defined scope.
What does a typical project cost?
MVPs and focused builds typically run $25k–$75k over 4–8 weeks. Production hardening, audits, and advisory engagements start lower; multi-month builds run higher. The fixed price comes from the discovery brief.
How long does a project take?
Most MVPs ship in 4–8 weeks. Audits and advisory engagements are 1–3 weeks. Larger production systems run 8–16 weeks. We commit to a date in the proposal and report against it weekly.
02
Team and process
Who actually does the work?
Senior engineers with shipped production AI experience. The same people on the discovery call write the code. No juniors, no offshore hand-offs, no account managers between you and the build.
How do you communicate during a project?
Shared Slack or Teams channel for day-to-day. A weekly written status update covering progress, risks, and decisions. A live demo at the end of every week. Async-first, no recurring meetings unless you want them.
What timezones do you cover?
Our team is based in Lahore (UTC+5). We have meaningful overlap with MENA, Europe, and the US East Coast. We schedule live calls in the client's timezone.
Will you sign an NDA?
Yes. Mutual NDAs are standard before discovery if your scoping call involves anything sensitive. We sign yours or send ours.
03
Technical, data, and IP
Who owns the code and IP?
You do. On delivery and final payment, all source code, models, prompts, and documentation produced under the engagement transfer to you. Standard reusable components (CI templates, eval harness scaffolding) remain ours but are licensed perpetually for your use.
Where does the code and infrastructure live?
Your choice. We can build in your GitHub org and your cloud account from day one, or in ours during development with a clean handoff at the end. We default to your accounts so there's no migration risk.
What about my data — does it leave my environment?
Only with your explicit consent and only when necessary. For training and evals, we use your data under your DPA. For LLM API calls in production, we configure providers with zero-retention settings where available (Anthropic, OpenAI, Bedrock, Azure OpenAI).
Do you train models on client data?
Never without an explicit, written agreement scoped to the specific engagement. We do not aggregate client data across projects, and we do not use any client data to train shared or open-source models.
What stack do you usually pick?
Defaults: Anthropic or OpenAI for LLMs; Next.js or FastAPI for serving; pgvector or Pinecone for retrieval; PyTorch + Ultralytics for vision; LangSmith or custom for evals. We pick what fits the problem — not what's trending.
Do you work with on-prem or air-gapped environments?
Yes. Open-source models (Llama, Mistral, Qwen) on your own GPUs, with vLLM or Ollama for serving and a private vector store. We've shipped systems that never touch a third-party API.
04
After launch
What happens after we ship?
You get a runbook, a dashboard, and a handoff session. From there you can take the system in-house, retain us for monthly iteration, or call us back when you need a new feature. No lock-in.
Do you offer support and SLAs?
Yes, on retainer. Standard tiers cover monitoring, bug fixes, and small feature work with response-time SLAs. Custom SLAs are available for production-critical systems.
What if my model degrades over time?
Drift is real. The eval harness we build runs nightly and flags regressions. On retainer we tune prompts, swap models, and re-train as needed. Without a retainer, we can run a one-off audit when you suspect degradation.
Still have questions?
Tell us about your project and we'll answer the rest. We respond within one business day.