This project is a conversational chatbot designed for Sonrisas de Bombay, an international NGO focused on fighting human trafficking and promoting human rights through education, awareness, and community empowerment.
The chatbot is currently live and integrated into the Sonrisas de Bombay website, where it supports users in real interaction contexts by answering frequent questions related to the online store, donations, and volunteering. Its main goal is to provide clear, accessible information while reducing friction and repetitive support queries.
You can try the chatbot directly on the Sonrisas de Bombay website: https://www.sonrisasdebombay.org
The chatbot is available in Spanish, reflecting the primary language of the NGO’s audience.
Sonrisas de Bombay serves a diverse audience with varying levels of digital literacy. Users often arrive with specific questions related to payments, shipping, returns, donations, or ways to collaborate, and expect quick, clear answers.
The main goals of the chatbot were:
- To improve access to key information without overwhelming users
- To guide users through common questions using natural, respectful language
- To reduce fallback responses by aligning the chatbot’s design with real user expectations
- To support the NGO’s mission through careful, ethical handling of sensitive topics
I designed, built, and iteratively improved the chatbot end-to-end.
My responsibilities included:
- Conversational design and intent architecture
- Training data creation and continuous optimization
- UX decisions based on real user interactions and no-match analysis
- Design and implementation of buttons, links, and custom payloads to guide users
- Ongoing refinement of fallback strategies to reduce user friction
Through continuous iteration and analysis of real usage, the chatbot evolved from an initial structure into a system currently managing 56 intents in production.
The chatbot was designed using an intent-based architecture, with a strong focus on clarity, accessibility, and guided interactions. Rather than relying solely on free-text input, the conversation design prioritizes buttons, quick replies, and structured responses to reduce cognitive load and help users move forward easily.
Custom payloads are used to:
- Present clear options when users are unsure how to proceed
- Redirect users to relevant sections of the website
- Reduce ambiguity in sensitive or complex topics
The intent structure is organized around real user needs (payments, shipping, donations, volunteering, general information), avoiding overly granular intents when a single well-designed response can cover multiple variations of a question.
The chatbot has evolved through continuous iteration based on real user interactions.
By analyzing conversation logs and no-match patterns, several recurring user behaviors were identified, leading to targeted UX improvements. These improvements significantly reduced friction and improved the perceived helpfulness of the chatbot without increasing complexity.
- Chatbot live and integrated into the Sonrisas de Bombay website
- 56 intents actively managed in production
- Use of buttons, links, and custom payloads to guide user interactions
- Continuous monitoring of conversation logs to identify improvement opportunities
- Ongoing refinement of training data and fallback strategies based on real usage
This project reinforced the importance of designing conversational systems around real user language and behavior, rather than assumed needs.
Key learnings include:
- The value of guided interactions over open-ended questions
- How small UX adjustments can significantly reduce fallback rates
- The importance of ethical language design in sensitive contexts
- The role of continuous iteration in maintaining conversational quality over time
Potential next steps include:
- Deeper analysis of user journeys across multiple intents
- Further refinement of proactive suggestions based on frequent user paths
- Exploration of additional accessibility improvements for diverse audiences
If you have any questions, feel free to connect with me on LinkedIn.