Frequently Asked Questions
Straight Answers Before You
Commit
Answers to the questions enterprise teams ask before starting an AI or software project with Bafar Labs — cost, timelines, data handling, intellectual property, deployment options and what happens after launch.
Working Together
- 01
What does Bafar Labs do?
- Bafar Labs is a global AI studio that builds production AI systems and the software around them. That covers voice agents, conversational AI, enterprise RAG, computer vision and custom model development, plus the web, mobile and platform engineering required to put those systems into daily operational use.
- 02
How does a project start?
- With a short discovery phase in the first days of the engagement. We map the workflow, audit the data and systems involved, and agree the metric that will decide whether the project worked. It ends with an architecture, a fixed scope and a fixed price — before any build commitment on either side.
- 03
Is there a minimum project size?
- In practice yes, because below a certain scope the discovery and integration overhead outweighs what can meaningfully be delivered. Our smallest engagement is a scoped pilot proving a single workflow. If what you need sits below that threshold we will say so in the first conversation rather than take on work that cannot succeed.
- 04
Can you work alongside our in-house engineering team?
- Yes, and it usually shortens the engagement. A common split has us owning the AI-specific layer — retrieval, evaluation, guardrails, model operations — while your team owns the application and domain logic. We work in your repositories, your sprint cadence and your tooling.
- 05
Which markets do you work in?
- We deliver to organisations across APAC, the Middle East, the Americas and Europe, working remotely with overlap hours agreed at the start of each engagement. Projects are quoted and invoiced in USD by default.
- 06
Which industries do you serve?
- Healthcare, financial services, retail and e-commerce, logistics, real estate, education, government and HR. The underlying AI techniques transfer between sectors; the domain workflows, regulatory constraints and integration surfaces do not, which is where discovery time is spent.
Cost & Timelines
- 01
How much does an AI project cost?
- Every project is quoted individually. Cost is driven by how many systems the solution has to integrate with, the state of your data, and whether the workload needs custom model work — not by a standard rate card. We fix scope and price after discovery, so the figure you receive is the figure you pay rather than an estimate that widens during the build.
- 02
How much does custom software development cost?
- Software builds are quoted individually after the definition phase, because cost is driven by the integration surface, the number of user roles, and the state of any existing system being extended. The delivery timeline is agreed in that same phase rather than published as a range, since a figure set before the integration surface is known would not survive contact with it.
- 03
How long does a project take?
- You see a working AI pilot on your own data in 14 days — discovery and architecture in the first few days, a functioning system by day fourteen. Beyond the pilot, production timelines are agreed during discovery rather than published as a range, because they depend on how many systems the solution has to integrate with, the state of your data, and your compliance requirements. Integration with legacy systems is the usual source of variance, not the AI itself.
- 04
Do you work fixed-price or time-and-materials?
- Fixed-price for defined scopes, which is most project work. The price is set after discovery, when the scope is genuinely understood, rather than estimated from a brief. Dedicated pods are monthly. We do not quote a fixed price before discovery, because a number produced at that stage is a guess presented as a commitment.
- 05
What are the ongoing running costs?
- Model inference, infrastructure and any third-party services such as telephony, all volume-dependent. We model these during discovery and report cost per conversation, per document or per call, so the unit economics are known before you commit. For most deployments, running costs are a small fraction of the build cost.
- 06
What are your payment terms?
- Fixed-scope projects are milestone-based, typically an initial payment at kick-off and the remainder against agreed delivery milestones. Retainers and pods are invoiced monthly in advance. Terms are set out in the contract before work begins.
Technical
- 01
Which AI models do you use?
- We select per workload rather than standardising on one provider — Claude, GPT, Gemini, Llama and Mistral are all in production use, chosen on accuracy, latency and cost for the specific task. Where a workload has mixed difficulty, we route easy requests to a smaller model and reserve the large one for hard cases.
- 02
How do you prevent AI hallucinations?
- Through architecture rather than instruction. Answers are grounded in retrieved sources and cited, structured outputs are validated against a schema and rejected when malformed, confidence thresholds route uncertain cases to a human, and an evaluation harness scores every change before it ships. Instructing a model not to invent things does not work; constraining what it can say does.
- 03
Do we need to fine-tune a model?
- Usually not. Fine-tuning changes how a model responds, not what it knows, so it is the wrong tool for supplying company knowledge — retrieval handles that better and stays current without retraining. Fine-tuning earns its place for consistent output structure, a specific tone, or narrow classification where prompting proves unreliable.
- 04
Can you deploy on-premise or in our own cloud?
- Yes. We deploy into your AWS, Azure or GCP account, into a private VPC, or fully on-premise. Open-weight models such as Llama and Mistral run entirely within your infrastructure, which is the standard route for organisations whose data residency requirements rule out hosted model APIs.
- 05
Can you integrate with our existing systems?
- Yes — CRM, ERP, telephony, ticketing, data warehouses and internal APIs. Integration is typically the largest portion of an enterprise project, and it is scoped explicitly during discovery rather than assumed, because reconciling identifiers, permissions and failure behaviour across legacy systems is where timelines actually go.
- 06
What if the AI is not accurate enough?
- That is what the prototype phase exists to determine. We measure accuracy, latency and cost against agreed criteria on your real data before production work starts. If the numbers do not clear the bar we say so, and either adjust the approach or recommend not proceeding — which costs far less than discovering it after a full build.
- 07
How do you measure whether an AI system is working?
- With an evaluation harness: a fixed set of representative inputs with known-good outputs, scored automatically on every change. It runs in CI, so a prompt or model change that degrades quality fails the build rather than reaching users. The harness is handed over with the system.
Data, Security & IP
- 01
Do you use our data to train models?
- No. Your data serves your application and nothing else. We deploy against provider endpoints with training explicitly disabled, and where that is insufficient for your compliance position we run open-weight models inside your own environment so no data leaves your perimeter at all.
- 02
Who owns the intellectual property?
- You do. All source code, prompts, fine-tuned weights, evaluation sets, infrastructure configuration and documentation produced for your project transfer to you on final payment. We retain no licence over your data or your models, and there is no proprietary framework you must keep paying to use.
- 03
Do you sign NDAs and data processing agreements?
- Yes, both as standard, and we complete enterprise security questionnaires as part of normal procurement. Subprocessor lists are available on request. Most enterprise security reviews complete without escalation.
- 04
Where is our data stored?
- In the region you specify. Data residency is treated as a hard constraint that shapes the architecture, not a preference — it determines which cloud regions and which model endpoints are eligible, and where hosted APIs cannot satisfy it, self-hosted models are used instead.
- 05
How do you handle personal data?
- PII is redacted at the logging boundary, so transcripts can be retained for evaluation and debugging without storing regulated personal data. Retrieval is permission-aware, meaning a user only ever receives answers drawn from documents they are entitled to see. Retention periods are configurable.
- 06
How do you protect against prompt injection?
- By limiting what a compromised model can do rather than trusting it not to be compromised. Tool permissions are scoped to the minimum the task requires, outputs are validated before any action is taken, and consequential operations require confirmation. A model that can only reply can only say something wrong; a model with unrestricted tool access can act on it.
After Launch
- 01
Do you provide support after launch?
- Yes, and AI systems need it more than conventional software because they drift as language, products and user behaviour change. Support covers quality monitoring against a held-out evaluation set, drift detection, model version upgrades and cost optimisation.
- 02
Can our own team operate the system?
- Yes — every build is designed for handover. You receive documentation, a runbook, the evaluation harness and a walkthrough with your engineers. A system only we can operate is a liability for you, so independence is the default rather than an upgrade.
- 03
What happens if we want to move to another vendor?
- You take everything with you. Code lives in your repositories throughout, infrastructure runs in your accounts where you prefer, and all documentation transfers on final payment. There is no proprietary runtime and no lock-in mechanism — if you leave, we would rather it be because the work is finished.
Looking for something more specific? The privacy policy sets out our legal obligations, the glossary defines the terminology, and each solution page carries product-specific questions.
Question Not Answered Here?
Ask it directly. We will give you a straight answer, including when the answer is that we are not the right team for what you need.