Summary
Pick A Voice’s is a voice bank and audiobook recording service connecting publishers and production teams with voice talent and studio production for their audiobooks. nexocode built Recordya, its own AI search and plugin engine, into the foundation of that platform, then delivered four domain-specific solutions on top of it: VoicePicker, an AI assistant that recommends the right narrator for a book; AI Market Scanner LIVE, an assistant that analyzes the Polish audiobook market in real time; an Orders view that surfaces live project data from Pipedrive; and a Knowledge Base that lets administrators share reference material across the platform. Because each tool is a plugin on a shared engine rather than a separate application, Pick A Voice’s’s users interact with all of them through one conversational interface.
The Problem
Finding the right voice for an audiobook is a high-stakes, largely manual decision. Production teams need to match a narrator’s tone, genre experience, and vocal character to a specific book, and the information needed to make that call — voice samples, portfolios, ratings, genre specializations — is scattered across spreadsheets and personal knowledge rather than searchable in one place. Getting it wrong costs time in re-recording or re-casting, and a slow casting process delays the whole production.
Alongside casting, publishers need to understand the market they are publishing into: which categories are growing on which platforms, which narrators show up most often in rankings, how competitors are performing. That data exists, but it is spread across external sources, changes constantly, and needs cleaning and reconciliation before anyone can act on it — work that, done by hand, doesn’t scale with the pace of a fast-moving market.
On top of casting and market intelligence, publishers needed a simple way to see the status of their own projects and orders, and Pick A Voice’s’s team needed a way to grant, extend, or revoke a publisher’s access to these tools without pulling in engineering for every change.
The Solution: Pick A Voice’s
Nexocode built Pick A Voices’s AI capabilities on Recordya, its agentic AI search and plugin engine, which supplies the base application architecture, a plugin mechanism, a chatbot/search layer, and a conversational user interface. Building on Recordya meant each Pick A Voice’s capability could ship as a domain-specific plugin on a common technical foundation, instead of as a separate, independently maintained application.
VoicePicker: AI-Assisted Voice Casting
VoicePicker is an AI assistant that recommends narrators for a given audiobook. A user describes the book — its genre, mood, the narrator’s expected gender or vocal tone, or any other preference — in plain language, and VoicePicker returns the best-matching candidates. The plugin reasons over data about each narrator: genre specialization, vocal character, audiobook experience, expert recommendations, ratings and reach, portfolio, voice demos, and photos, all presented as candidate cards.
Much of the implementation work went into preparing this data so the AI could use it safely. Narrator data is synced from Airtable, which serves as the operational source of truth, and the sync pipeline cleans and normalizes it — standardizing text fields, numbers, ratings, value lists, demo links, photos, and details added by the Pick A Voice’s team. As a result, a user doesn’t need to know the underlying data structure or field names — a request like “a male narrator for a crime novel with a low voice” is enough for VoicePicker to turn that intent into a safe, structured search and return concrete recommendations.
AI Market Scanner LIVE: Real-Time Market Intelligence
AI Market Scanner LIVE is a plugin for analyzing the Polish audiobook market. Users can ask about popular titles, categories, platforms, authors, publishers, and narrators, or about market trends generally. The plugin connects to an external market data source, from which information about audiobooks, rankings, platforms, categories, and market participants is pulled through an ETL process: fetched in batches, transformed, mapped to a common data model, and stored in a structured analytical database.
A significant part of the work was normalizing data coming from different platforms, since category names and the names of authors, publishers, and narrators can vary in format, spelling, or structure from one source to the next. nexocode built a data-cleansing process that maps platform-specific categories to a shared taxonomy, standardizes author, publisher, and narrator names, deduplicates records, unifies platforms and rankings, and syncs incrementally with watermarks so reprocessing the same data doesn’t create duplicates or drift. With that foundation in place, AI Market Scanner LIVE can answer questions like “which genres dominate on Storytel?”, “compare Audioteka and BookBeat”, or “which narrators appear most often in the rankings” without anyone opening a spreadsheet.
Orders View: Live Project Status from Pipedrive
The Orders view lets Pick A Voice’s’s users track the projects tied to their organization, with data pulled directly from Pipedrive, so nothing has to be re-entered by hand. It groups projects by organization, automatically finds or creates the right Pipedrive filter for each organization, pulls current project data through the Pipedrive API, maps Pipedrive-specific fields such as project phase, assigned narrators, and links, and refreshes automatically via webhooks when something changes in the CRM. Publishers get a single place to check the status of their projects, while Pipedrive stays the system of record for Pick A Voice’s’s sales and operations team.
Knowledge Base: A Shared Resource Library
The Knowledge Base is a lightweight library of links and materials available to all Pick A Voice’s users. An administrator can add, edit, or remove entries, and logged-in users can browse what’s been published — instructions, process documentation, onboarding material, resource links, or other reference content — without engineering needing to be involved in every update.
Access Management, Without Engineering Involvement
Pick A Voice’s’s administrators can grant, extend, or revoke a user’s access to individual products — VoicePicker and AI Market Scanner LIVE — directly from the application, without touching the database. Access is managed per user and per product: an admin can set an expiration date, extend access by a given number of days, or revoke it by setting it to expire. If a user was created outside the standard Pipedrive-driven flow, the application can create a technical profile for them so their access can still be managed the same way. This cuts down on manual admin work and lets Pick A Voice’s turn on access to a feature quickly when a customer needs it.
A Conversational Interface That Talks Back in Widgets
Recordya’s chatbot/search layer means Pick A Voice’s’s users interact with all of these tools through one conversational interface, not several separate ones. The AI reads the user’s intent, works only against a defined, safe set of read operations, and uses domain-specific prompts to keep responses short, business-focused, and understandable to non-technical users. When a question is ambiguous, the interface can generate a clarifying widget — a set of clickable options — instead of a back-and-forth in plain text.
When there’s an answer, results are rendered as whichever UI component fits them best: narrator cards, tables, lists, action elements, or media players for voice demos. In VoicePicker, that means candidate cards with a narrator’s key details, recommendations, photo, portfolio link, and a way to play their demo; in AI Market Scanner LIVE, it means tables and structured summaries alongside a short written interpretation. VoicePicker’s language handling also accounts for Polish-specific grammar, such as the adjective inflections used to describe vocal tone. This behavior is covered by evaluation test sets.
Impact
The project delivered working AI plugins and operational views that now support Pick A Voice’s’s day-to-day work:
- Casting a narrator got faster, since VoicePicker turns a plain-language description into a ranked shortlist instead of a manual search through spreadsheets and personal contacts.
- Narrator and market data is now clean, structured, and kept in sync automatically from Airtable and external market sources, instead of living in scattered files.
- Publishers and production staff can ask questions in natural language — about a narrator or about the market — instead of building filters or reading raw data exports.
- The Orders view gives publishers a live, self-service view of their project status, pulled directly from Pipedrive.
- The Knowledge Base gives administrators a way to publish and update reference material without involving engineering.
- Access to VoicePicker and AI Market Scanner LIVE can be granted, extended, or revoked per user and per product directly from the app.
- Building on Recordya meant Pick A Voice’s reused a proven chatbot, search, and plugin foundation instead of building separate applications, and got a conversational interface that renders results as cards, tables, and other widgets rather than plain text.
- The work sets up a foundation for adding further Pick A Voice’s plugins and automations on the same Recordya-based architecture.
