The answers your review team will ask for.

We provide human-in-the-loop AI evaluation and customer-experience quality operations. Typical work includes agent-response review, RAG answer and retrieval review, conversation QA, handoff and escalation audits, taxonomy design, calibration, and release evidence.

We do not claim ownership of your models, deploy changes into your production environment, or represent illustrative pilot examples as client work. Legal, regulatory, and production release decisions remain with the customer unless a written agreement states otherwise.

After a scoped kickoff, we align on sample selection, scoring criteria, acceptance thresholds, escalation paths, and reporting. The review team then evaluates an agreed sample, calibrates against a shared reference set, and returns a decision-ready report.

Depending on scope, deliverables can include a calibrated rubric, reviewed sample set, error taxonomy, issue log with evidence, severity and ownership recommendations, reviewer agreement notes, and a concise readout of release or improvement priorities.

We can operate a recurring quality pod, a throughput-based review program, or a defined project workstream. Each engagement starts with an agreed cadence, quality target, response expectation, and named delivery owner.

Before production work begins, we define the rubric, label definitions, acceptance criteria, and escalation rules with the customer. Reviewers calibrate on shared examples; disagreement and edge cases are documented and fed back into the operating guide.

Yes. Customer policy, brand voice, safety rules, and product-specific requirements are converted into a usable review guide. We identify ambiguous cases early rather than silently inventing a rule.

The default delivery posture is a customer-hosted workspace, role-based access, documented QA, and no unapproved subcontracting. We minimize access to the material required for the agreed task and record the delivery process in the statement of work.

Only by explicit agreement. Before production data is made available, the customer and Blueheaven confirm the approved workspace, data categories, access roles, handling rules, retention expectations, and escalation contacts in the engagement documents.

Yes. We provide a scoped operating description for vendor review and work through reasonable security, data-processing, and onboarding questions before delivery begins.

A pilot is normally quoted as a fixed scope after we confirm review volume, languages, data sensitivity, rubric maturity, turnaround expectation, and reporting requirements. Scope and price are confirmed in writing before work starts.

Common models are a monthly managed quality pod, agreed throughput by review unit, or a project statement of work. The right model depends on volume variability, operating hours, language coverage, calibration needs, and the customer’s preferred procurement structure.

Blueheaven Nepal is wholly owned by Blueheaven (Shenzhen) Technology Co., Ltd. The contracting entity, data-processing roles, confidentiality terms, and ownership of data and agreed deliverables are specified in the MSA, SOW, and any required DPA before delivery begins.