Visit Website: www.practivated.com
The platform featured four predefined subscription tiers designed to meet different user needs. Basic plans offered limited usage, such as 20–25 minutes per week, while enterprise-level subscriptions provided fully customizable options tailored for large organizations with specific requirements.
A "minutes pool" mechanism was introduced to support team-based usage. Administrators could allocate conversation time credits to team members and manage user licenses efficiently, ensuring flexibility and control over resource distribution.
User roles were clearly defined, including admin, manager, and regular user permissions. This role-based system enhanced data security by restricting access to sensitive features and ensuring only authorized users could manage subscriptions or allocate resources.
To overcome this, we implemented a dynamic prompt approach, allowing users to input information directly into the prompt to create a personalized donor persona. Users could provide a website link (for individual donors or organizations), which we would scrape for relevant data, summarize, and dynamically incorporate into the persona creation process. This ensured the persona was tailored and well-informed.
We have implemented an approach to seamlessly resume conversations after a pause. This is achieved by summarizing the previous conversation and passing the summarized format forward, ensuring the context of the conversation is preserved and continuity is maintained.
We built a modular subscription system to address these needs. This system included adjustable subscription tiers, a "minutes pool" feature for team usage, and role-based permissions to ensure that administrators could manage resources and licenses effectively without compromising user experience or security
Iterative refinement of prompts using the RAG model was undertaken to fine-tune feedback content. Furthermore, the client’s voice was trained using 11 Labs, enabling the delivery of audio feedback that matched her style and tone, making the user experience more engaging and natural.
A robust preprocessing pipeline was implemented to handle the extracted data's normalization, deduplication, and structuring. OpenAI models were then used to generalize patterns from this diverse input, ensuring that donor simulations were both realistic and adaptable to varied organizational contexts.
The platform heavily relies on services like VAPI and OpenAI for critical functionalities, which can introduce latency issues and increase operational costs.
Currently supports only English, limiting the platform’s usability in non-English speaking regions and global markets.
The platform’s ability to assess soft skills is limited due to its reliance on textual and audio data, which cannot fully capture interpersonal dynamics.
Web crawling and AI training must strictly adhere to data privacy regulations and platform terms of service, which can be restrictive and challenging to navigate.
Node.js: Real-time interaction and API management.
MongoDB: NoSQL database for storing dynamic JSON data such as conversations and user interactions.
Hosted on AWS S3 for scalable storage solutions.
VAPI: Enabled real-time audio-to-audio interaction between users and the AI donor.
11 Labs: Trained client’s voice for synthesized audio feedback.
OpenAI: Powered chatbot responses and contextual feedback generation.
Google authentication and OAuth for secure user management.
Byte.js: Designed for lightweight, responsive user interfaces with functionality as the priority.
Python scripts for extracting relevant donor organization data.
