How to read this register
This register lists the providers BuildInterview uses. Not every provider receives every candidate’s data: an interview’s AI assistant answers with the model the hiring company chose for it, a small OpenAI model checks each message a candidate sends the assistant (never the files) before it answers, and the fallback reviewers are used only when the primary reviewer cannot complete a review.
The customer’s agreed processing scope must identify the actual providers and features used for its workspace. Provider names here identify services; the applicable contract establishes the contracting legal entities, processing locations, safeguards and any subprocessor-notice rights. Provider documentation is linked for transparency and does not itself prove that a particular account setting or agreement is in place.
Infrastructure and delivery
Amazon Web Services
Data privacyPrivate evidence and recording storage, encryption, queues, and the workers that process recordings and run automated evaluation.
Data: Assessment files, recordings, evidence, processing jobs and operational logs, depending on the service.
Core infrastructure.Supabase
Data-processing termsCompany authentication, database records and organization/assessment authorization.
Data: Account and organization information, candidate and invitation records, assessment state, usage, audit records and stored assessment data.
Core identity and database services.Vercel
Data-processing termsWebsite/application hosting, server requests, delivery and deployment infrastructure.
Data: Network and request information, operational logs and information handled by the application endpoints.
Application hosting.Cloudflare
Privacy and TurnstileDomain/DNS services and configured security services, including Turnstile bot protection.
Data: Network, device and browser/security signals; traffic processed depends on the configured service.
Domain and security services. Turnstile also describes its own processing in its privacy notice.Resend
Data-processing termsAuthentication and transactional email delivery, including invitations and account updates.
Data: Recipient email, message content, private links needed in a message, and delivery or failure events.
Transactional email. Do not forward private invitation messages.Isolated execution environments for candidate code and separate grading workloads.
Data: Question-authorized files and code, execution input/output, grading material within its separate environment and related runtime metadata.
Configured server execution. Public browser demos may instead run locally.
Browser asset delivery
The interview room’s camera checks download three public Google MediaPipe models from Google-hosted storage (a face landmark model, a face detector and a person detector) and their runtime assets from jsDelivr. These requests disclose ordinary connection information, such as your IP address and browser request headers, to the relevant delivery service. The checks run on the candidate’s device and do not upload camera frames to those asset hosts.
These are public software downloads, separate from AI assistance, submission evaluation and recording storage. See Google Cloud’s privacy notice and jsDelivr’s privacy policy. The delivery providers and their network locations should also be considered when a customer requires restricted processing regions.
Website analytics
These run only on the public website, never in the company workspace, on candidate invitation links or in the interview room. In the EEA, the UK and Switzerland they wait for the visitor to accept; until then Google Analytics and Vercel only count visits without cookies. Visitors can turn them all off in Cookie settings.
Google (Google Analytics)
Data-processing termsCounts visits to the public website and shows which pages and links help visitors, and how they arrived.
Data: Pages viewed, referrer and campaign tags, device and browser, approximate location from the network address, clicks on links such as Create account, and a pseudonymous cookie identifier.
Public website only. No advertising features or Google signals. Event data is kept for 14 months.Microsoft (Clarity)
Clarity termsHeatmaps and replays of how visitors use the public website, so we can fix pages where people get stuck.
Data: Clicks, scrolling and page layout as a replay with typed text hidden, device and browser, approximate location and a pseudonymous cookie identifier.
Public website only. Recordings are kept for 30 days.Vercel (Web Analytics)
Web Analytics privacyCounts visits to the public website without cookies.
Data: Pages viewed, referrer, device and browser, and country from the network address. No cookies and nothing stored in the browser.
Public website only. Stops when a visitor turns analytics off.Cloudflare (Web Analytics)
Web AnalyticsCounts visits to the public website and measures how quickly its pages load for real visitors, without cookies.
Data: Pages viewed, referrer, device and browser, country from the network address and page-load timings. No cookies.
Public website only, after consent where it is required.
Payments
Razorpay
Privacy informationCompany checkout, payment processing and payment confirmation for BuildInterview purchases under AIDriven Technologies Private Limited.
Data: Order reference, amount, currency and payment status. Razorpay collects the payer and payment-method information required by its checkout directly; we do not collect full card details.
Applies when Razorpay checkout is enabled. Assessment answers, recordings and candidate reports are not sent for payment processing. Razorpay also processes information for its own payment, fraud-prevention and legal obligations.
AI providers
Dictation is a third, separate workload: when a candidate chooses the microphone in an Audition, only that clip is sent to Google’s Gemini API for transcription, or to OpenAI if Google is unavailable. It transcribes without rewriting, does not use the AI allowance, and the audio is not stored by BuildInterview.
Candidate assistance and submission evaluation are separate workloads. The assistant in an Audition answers with the model the hiring company chose for that interview. Before it answers, a small OpenAI model (GPT-6 Luna) checks each message the candidate sends it, never the files, to decide whether the assistant may write code for it: it writes code for a part the candidate describes in their own words or picks from the assistant’s own list. The assistant has no internet access and no tools, and it sees only that Audition’s brief, files and the candidate’s messages. Selecting “No AI” assistance does not remove automated evaluation: submitted work is reviewed by our AI reviewer, which may use a different provider from the assistant.
OpenAI
API data controlsCandidate AI assistance when the hiring company chooses an OpenAI model, the message check in every Audition (a small model, GPT-6 Luna, reads each message a candidate sends the assistant, never their files, and decides whether the assistant may write code for it), the primary AI reviewer of submitted work, and the fallback transcription of a candidate’s dictation when Google is unavailable.
Data: The candidate’s messages with the Audition context the assistant needs (the brief, file contents and dataset schemas) and its responses; for the message check, the candidate’s latest message, a few recent turns of the chat with code left out, and any error text the candidate pasted; for review, the submitted work, plan, writeups and evaluation instructions; for dictation, the audio clip the candidate chose to record and its transcript; usage records.
Direct API. Requests carry no tools and ask OpenAI not to store them; abuse-monitoring retention follows OpenAI’s API terms.Anthropic
Commercial data useCandidate AI assistance when the hiring company chooses a Claude model, the first fallback reviewer if the primary reviewer cannot complete a review, and the workspace assistant (Ask) for signed-in members of the hiring company.
Data: The candidate’s messages with the Audition context the assistant needs and its responses; for a fallback review, the material the primary reviewer would have read; for the workspace assistant, the member’s questions, the workspace facts it looks up for them (such as roles, stages, scores and dates, with each candidate shown as a stand-in instead of their name, and never an email address), the text of any job description they share, and its replies; usage records.
Direct Claude API, which Anthropic does not use for training. Candidate and reviewer requests carry no tools. The workspace assistant looks things up with the member’s own access, stays on the model we name and changes nothing until the member confirms. Conversations are kept in BuildInterview’s own database, not with Anthropic.Google (Gemini API)
API service termsTranscription of a candidate’s optional dictation in an Audition (Gemini 3.5 Transcribe), and candidate AI assistance when the hiring company chooses a Gemini model.
Data: For dictation, the audio clip the candidate chose to record and its transcript; for the assistant, the candidate’s messages with the Audition context it needs and its responses; usage records.
Paid Gemini API service, whose terms exclude prompts from product improvement. Requests carry no tools and no Google Search grounding, and dictation requests ask Google not to store them.DeepInfra
Privacy informationThe second fallback reviewer, used only when neither the primary nor the first fallback reviewer can complete a review.
Data: The material the primary reviewer would have read, the review output and usage records.
Direct API for a hosted open-weight model. DeepInfra documents that inputs and outputs are not stored or used for training.
Processing locations and transfers
Information may be processed in the United States and other countries where the relevant provider or authorized support operation works. A cloud storage region does not describe every email, support, edge-delivery or AI-processing location.
Confirm the actual deployment regions and any model’s cross-region routing before sending data that has residency restrictions. Where required, the customer agreement must include appropriate transfer safeguards and a service-specific processing schedule. We do not promise that all data stays in India, the EU or another single jurisdiction.
Training, retention and minimization
AI requests should contain only the context needed for the task. Avoid production secrets and unnecessary personal or sensitive information in company questions, datasets and prompts.
Provider training rules and retention rules are separate. A service can exclude customer data from model training while still retaining requests for abuse monitoring or legal obligations. BuildInterview does not make a blanket zero-retention claim across these services.
We review each AI provider’s data terms, paid-service status, regions and deletion arrangements before enabling it for candidates. The workspace’s evidence-retention policy does not automatically change a provider’s retention terms.
Provider changes and questions
We update this register when the provider scope changes. Where a data-processing agreement requires advance notice, authorization or an opportunity to object, that agreement governs the change process. Contact us to request the applicable provider schedule or raise an objection under your agreement.
For individual data requests, see the privacy notice. Candidates should also contact the hiring company, which controls the assessment’s purpose and use.
Contact AIDriven Technologies Private Limited, the operator of BuildInterview in Delhi, India, at hello@buildinterview.com for subprocessor, transfer or data-processing questions. Include the relevant page or hiring company and a short description. Do not email passwords, invitation links, identity documents, medical records or candidate recordings. We will arrange an appropriate channel if more information is needed.
BuildInterview is operated by AIDriven Technologies Private Limited.
Back to top ↑