• ✓ You have a job listing page — a page where you post call center jobs, CSR openings, or any BPO vacancy.
• ✓ We give you one <script> tag. You paste it into the HTML of that page before the closing </body>.
• ✓ The CONNECT widget loads inside a Shadow DOM — it cannot break your existing page styling or layout.
• ✓ Done. Candidates can now submit 45-second video profiles and be scored by AIVA before you ever look at them.

• ✓ You have any website — your BPO company website, your job board, your HR portal. Any page.
• ✓ We give you one <script> tag. You paste it once. Maya appears as a chat panel on that page.
• ✓ Maya speaks 30 languages and handles candidate and employer questions out of the box, with no additional configuration on your part.
• ✓ Done. Maya is live on your site. Text channels (website chat) are free forever — no monthly fee for the chat widget itself.

• 1 You register a Business account on the platform you want to use — WhatsApp Business API (Meta), Viber Business (partners.viber.com), or LINE Official Account (line.biz). This is your account. We do not register it for you.
• 2 You provide us with the API credentials — channel access token, channel secret, webhook verify token, and phone number or account ID. These differ slightly by platform.
• 3 We register the webhook URL — pointing your messaging platform account at the Maya API endpoint that handles inbound messages from your platform.
• 4 We configure phone number routing (WhatsApp and Viber) — linking your business number to Maya so messages from your customers arrive at Maya and responses go back out through your account.
• 5 We test the connection end-to-end — send a test message, verify Maya responds, confirm language detection is working correctly.

• 1 You provide a phone number — either a number you already own (which we port to Retell AI) or a new number we provision for you through Retell AI. This is the number your customers will call or that Maya will call out from.
• 2 We set up the Retell AI agent — connecting your phone number to Maya, configuring the AI voice (Azure TTS with your language — fil-PH, th-TH, en-US, or any of 140+ languages), and setting the speech-to-text language (Deepgram).
• 3 We configure Caller Choice — on every call, the caller is offered the choice of AI or a human agent at the start of the call. You specify the default and whether you want the choice offered. We build the call opening flow to your specification.
• 4 We set up call routing — defining how calls are handled: inbound only, outbound only, or both. If you are using Maya to replace L1 agents, we configure the escalation path to a human for calls Maya cannot resolve.
• 5 We run test calls — inbound and outbound, verify the voice quality, confirm Caller Choice is working, confirm Maya's responses are appropriate for your use case.

RAG stands for Retrieval-Augmented Generation. In plain English: instead of Maya inventing an answer from general training, Maya looks up the answer in your actual documents first, then responds based on what it finds. The answer comes from your call scripts, your SOPs, your training manuals — not from the AI's general knowledge.

This requires real web development work. Here is what is involved:

• 1 You supply your data. This means sending us the actual documents you want Maya to search: call scripts, escalation guides, HR policy PDFs, FAQ documents, training manuals, SOP documents, product catalogs, legal contracts — whatever is relevant to the questions your team or your customers ask. We cannot build this without your data. There is no default content.
• 2 We prepare the data. Documents are cleaned, structured, and split into searchable chunks. PDFs are parsed, tables are extracted, inconsistent formatting is corrected. This step varies in time depending on the quality and volume of what you supply.
• 3 We vectorize the data. Every chunk is converted into a 1,536-dimension vector embedding using Voyage AI. This is the process that makes semantic search possible — Maya can find the right answer even when the question is phrased differently from how the document is written.
• 4 We build the search interface. A search endpoint is configured specifically for your corpus — your data, your document types, your use case. For a BPO client, this might be a call center agent search bar that returns answers from your escalation guide in real time. For a different client, it might be an internal HR policy search. The interface is built to your specification.
• 5 We connect it to Maya. Once the search engine is running, Maya's responses are augmented with it — every answer Maya gives on your channels (website chat, WhatsApp, phone) checks your document store first before responding. This is the step that eliminates hallucinated answers.
• 6 We test against real queries. We run representative questions from your actual use case, verify the correct chunks are being retrieved, and tune the search threshold. You review the output and confirm it meets your standard before we go live.