Context-aware AI assistance for real conversations.
Send the conversation. Understand the context. Get three ways to reply.
Desenrola is an AI-assisted conversation product designed to reduce the friction of deciding what to say next.
Instead of treating every message as an isolated prompt, the product works around people, conversation history and persistent context.
A user can provide a conversation, receive multiple reply directions, refine the result and keep relevant context associated with each contact.
The goal is not simply to generate text.
The goal is to make the next response fit the conversation, the person and the user sending it.
Messaging tools usually start from zero.
Real conversations do not.
Who the person is, what happened before, the current tone, previous interactions and the user's own communication style can all affect what makes sense as the next reply.
Desenrola was built around that gap.
Instead of repeatedly explaining the same situation, context can remain associated with the person and inform future interactions.
A typical flow looks like this:
1. Add the conversation
The user provides a screenshot or conversation context.
2. Understand the situation
Desenrola processes the available context for that specific interaction.
3. Generate three directions
Instead of returning one definitive answer, the product offers three possible replies.
4. Refine
The user can adjust the direction, tone or provide additional information without restarting the process.
5. Keep the context
Relevant history and information remain associated with that contact for future conversations.
One message can have several valid answers.
Desenrola generates multiple options so the user can choose a direction instead of being locked into one generated response.
If a response is close but not right, it can be refined rather than regenerated from zero.
The user can change tone, direction or add new context.
Each person has an individual context instead of sharing one generic conversation memory.
This allows information and history to remain tied to the correct contact.
Previous interactions can remain available as useful context for later conversations.
Desenrola can also use information about the user, helping generated responses move closer to the way that person would naturally communicate.
The product tracks how long a contact has been inactive and uses a simple visual state system to help surface conversations that may be cooling down.
A relationship can span different messaging platforms while still belonging to the same underlying contact context.
Desenrola is built as a full-stack web product combining application logic, persistent data, authentication, AI infrastructure and payments.
React · Next.js
Responsible for the web application experience and product interface.
Node.js · Python
Used across server-side application logic and supporting backend workflows.
Supabase
- database;
- authentication;
- file storage.
Vercel
- application hosting;
- deployment infrastructure;
- AI Gateway.
Anthropic Claude · Vercel AI Gateway
AI capabilities are integrated into the product workflow rather than exposed as a generic chatbot interface.
Mercado Pago
Used for the product's payment infrastructure.
Conversations rarely have one objectively correct answer.
Providing multiple directions turns generation into a choice and gives the user more control over how they want to respond.
Re-explaining the same person every time creates friction and produces less useful answers.
Keeping context associated with each contact allows future interactions to begin with more relevant information.
Generated text is often almost right.
Refinement lets the user move an answer toward the desired tone or intention without discarding the useful parts.
A response can make sense logically and still feel unnatural for the person sending it.
Personal context helps bridge that gap.
I co-build Desenrola across software development, product decisions, UX and practical AI integration.
My involvement extends beyond isolated implementation tasks: the product is developed through continuous iteration across engineering, interface decisions, feature behavior and the way AI is integrated into the user experience.
Desenrola is built together with Allan Hert.
Desenrola is an actively developed product.
Its production source code, internal infrastructure details, credentials and proprietary implementation remain private.
This repository is the public product and engineering case study for the project.
It is intended to document the product, its public architecture, design decisions and the work behind it without exposing proprietary source code.
React · Next.js · Node.js · Python · Supabase · Vercel · Anthropic Claude · Mercado Pago
Live product: https://desenrola.app/
André Miller Fleck: https://github.com/DredsGG
LinkedIn: https://www.linkedin.com/in/andrefleckm/
Built in Brazil by Allan Hert & André Miller Fleck.