Powered Virtual Assistant For Ideation Sessions

Waverley worked with the client through an iterative R&D process, pivoting from the smart speaker vision toward a web-based conversational platform that could deliver more immediate value while establishing a foundation for future hardware deployment.

Architecture and Infrastructure

Cloud Platform: Google Cloud hosts the application, providing scalable infrastructure and access to advanced speech and NLP services.

Container Orchestration: Kubernetes manages containerized services in production, while Docker provides local development containerization.

Backend Stack: TypeScript with NestJS framework provides the backend—chosen for fast execution and as a strong alternative to Java Spring in the JavaScript ecosystem.

Frontend: React enables a responsive, intuitive user interface for managing conversations, ideas, and collaboration.

Database: MySQL stores conversation history, idea data, form submissions, and user collaboration metadata.

ML/NLP Services: Python-based microservices handle machine learning and natural language processing tasks, leveraging scikit-learn, NLTK, and FastText.

Chatbot

Chatbot with voice recognition and synthesis ability. This feature is implemented with the help of Google Speech-to-Text and Text-to-Speech services in conjunction with our custom back-end algorithm for improved pause detection and continuous dictation ability. This was an effective solution to the problem of speech length restriction and poor pause detection. Now, the user may speak for an unlimited time, make pauses to take a breath, and finish their thoughts. Meanwhile, the app will listen and convert it all to text (with mostly accurate punctuation).

Core Features

Voice-Enabled Chatbot: Integrates Google's Speech-to-Text and Text-to-Speech services combined with custom backend algorithms enabling unlimited speech duration and improved pause recognition. Users can speak naturally, pause to collect thoughts, and the system accurately transcribes with punctuation.

Open-Domain Understanding: A critical challenge was enabling the system to understand and respond meaningfully to queries about arbitrary topics. Waverley's solution employs machine learning algorithms that analyze queries, classify knowledge domains, retrieve relevant information from the Wikipedia database using semantic similarity search, and identify frequently-used associations to provide relevant responses.

Creativity Tools and Brainstorming Methods: The system implements established brainstorming methodologies including association-based ideation and Osborn's Checklist (SCAMPER). The ML system generates both random and targeted associations to stimulate creative thinking, recognizing that both types play important roles in the ideation process.

Intelligent Conversation Flow: A business analyst worked with Waverley engineers and domain experts (experienced ideation facilitators) to design natural conversation flows. The system provides subtle, organic guidance rather than mechanical prompts, smoothly transitioning between brainstorming techniques while maintaining conversational naturalness.

Idea Management and Collaboration: Users can save ideas, view conversation history for each concept, and grant organizational colleagues access to specific ideas for collaborative refinement. Role-based access control (organization admin, ideation manager, inventor) enables fine-grained permission management.

Patent Integration: The platform automatically populates patent submission forms with relevant idea information, including title, description, business value, and novelty assessment. Integrated Google Patent Search enables users to identify existing patents related to their ideas.

Machine Learning Components

Information Retrieval: The system analyzes incoming queries to identify the relevant knowledge domain and retrieves top Wikipedia articles using semantic similarity. By analyzing word frequency and associations across retrieved articles, the system provides users with relevant concepts and associations for their query.

Binary Text Classification: To guide ideation effectively, the system classifies user responses as affirmative or negative, enabling sentiment-aware conversation flow and helping the chatbot understand whether the discussion is moving in productive directions. This capability enables the system to adapt its guidance in real-time.