Conversational Companion Milo
Conversational Companion Milo
Project Overview
Milo is a conversational wellness companion designed to help users navigate moments of stress, distraction, low energy, and emotional overwhelm through lightweight, emotionally aware interactions. The project explored how conversational UX could reduce friction during vulnerable moments by replacing complex wellness interfaces with calm, guided dialogue.
The experience lives within the Mindgrid ecosystem and focuses on helping users regulate themselves in the moment through breathing exercises, grounding techniques, movement prompts, and reflective check-ins. Instead of functioning like a productivity tool or a clinical assistant, Milo was intentionally designed to feel warm, reassuring, and human.
This project included conversational scripting, flow mapping, intent architecture, interaction design, and implementation using AWS Lex and AWS Lambda.
Duration: 7 Weeks
Team: Keertana, Rutuja, Kshitija
Role: My contributions included, Conversational flow design, Emotional tone and response writing, Intent and slot planning, Flow architecture, Conversational scripting, Reflection and conversational pacing design, Visual presentation and storytelling
Tools: Figma, AWS Lex, AWS Lambda, Flow Mapping, Conversational Scripting
The Problem
Emotional support tools often require too much effort during moments of stress
We began the project by thinking about the “in-between” emotional moments people experience throughout the day, moments where stress exists, but not necessarily at a crisis level.
These moments are often overlooked by wellness products. Most existing tools assume users have the time, energy, and emotional bandwidth to actively seek support. In reality, when someone feels overwhelmed, distracted, mentally drained, or emotionally exhausted, even opening multiple apps and navigating interfaces can feel burdensome.
We noticed that people often rely on several disconnected tools to manage emotional wellbeing like Meditation apps, Breathing exercises, Focus timers, Movement prompts, Journaling tools, Mindfulness content.
Switching between multiple interfaces creates friction precisely when users are least equipped to handle it.
This led us to ask:
Why Conversation Was the Right Medium
Before designing Milo, we evaluated whether conversation was actually the right interaction model for the problem space.
We found that people already naturally describe emotional states conversationally:
“I’m overwhelmed.”
“I can’t focus.”
“I feel low energy.”
“I’m distracted.”
Because emotional check-ins already exist in spoken language, a conversational interface felt intuitive and emotionally familiar.
Conversation also offered several advantages reduced screen navigation, lower cognitive effort, faster interaction during stressful moments, more emotionally responsive communication and lightweight guidance without overwhelming users.
Rather than forcing users through multiple menus and decision-heavy interfaces, Milo could simply begin with:
This became one of the project’s core interaction principles:
Design Goal
Create a conversational system that feels emotionally supportive without becoming overwhelming, robotic, or clinical.
From the beginning, we knew Milo could not sound like:
❌ A productivity assistant
❌ A medical chatbot
❌ A command-driven system
❌ An emotionally detached interface
Instead, we wanted the experience to feel:
A major design principle emerged early in the project:
This shaped almost every conversational decision that followed.
Understanding the User Context
One of the biggest insights during our early exploration was realizing that users interacting with Milo would often already be emotionally depleted.
This meant the system needed to account for:
✅ Reduced Focus, Low Patience, Emotional Fatigue, Cognitive Overload, Decision Fatigue
Traditional wellness interfaces often require users to Browse options, Choose programs, Read long explanations, Navigate multiple screens
During moments of stress, these interactions can unintentionally create even more friction.
Because of this, Milo was intentionally designed around: Short conversational prompts, Minimal decision-making, Guided interactions, Clear next steps and Low cognitive effort
Rather than giving users dozens of wellness options, Milo introduces one small actionable step at a time.
Emotional Framing
Designing for “in-between moments”
A major conceptual decision was defining what Milo was and what it was not.
Milo was not designed for:
❌ Crisis intervention
❌ Therapy replacement
❌ Diagnosis
❌ Intensive mental health treatment
Instead, Milo focused on everyday emotional regulation:
✅ Stress before work
✅ Low energy during the day
✅ Difficulty focusing
✅ Moments of overwhelm
✅ Emotional fatigue
✅ Nervous system dysregulation
We intentionally framed Milo around the small emotional moments that accumulate throughout daily life.
This positioning helped us avoid designing the experience as either:
➡️ Overly clinical
➡️ Or unrealistically therapeutic
Instead, Milo became a lightweight emotional support system focused on immediate relief and grounding.
Designing Milo’s Personality
Tone became one of the most important parts of the experience
A major part of conversational UX is understanding that wording changes emotional perception.
Because Milo exists in emotionally sensitive moments, tone design became central to the project.
We intentionally avoided any robotic language, productivity framing, urgent commands, emotionally cold responses, overly enthusiastic wellness language
Instead, Milo was designed to sound gentle, calm, validating, conversational and emotionally aware
For example:
Supportive Tone
✅ “That sounds really heavy, let’s try a simple breathing exercise together.” (Image)
Tone We Avoided
❌ “You are experiencing distress. Begin breathing exercise now.” (Image)
This distinction may seem small, but during testing and scripting, we realized these tonal differences dramatically changed how emotionally safe the interaction felt.
The goal was never to imitate a human therapist. Instead, the goal was to create interactions that reduced emotional friction.
Defining Emotional States
To simplify the interaction model while still allowing personalization, we organized the experience around four primary emotional states:
Stressed
Distracted
Low Energy
Calm
Each state triggered different conversational pathways and intervention techniques.
This structure helped Milo feel adaptive without becoming overly complex.
Instead of overwhelming users with many emotional categories, the system focused on broad emotional states that users could quickly recognize and select.
Flow Architecture
The system was designed as a multi-turn conversation rather than a one-time interaction.
The overall structure followed this pattern:
Check-In → Emotional State → Technique → Reflection → Retry → Close
This loop allowed Milo to:
Respond dynamically
Adapt based on feedback
Encourage lightweight reflection
Avoid abrupt session endings
We intentionally structured the experience as guided conversation instead of a decision-heavy interface.
This helped reduce cognitive effort during emotionally vulnerable moments.
Intent & Slot Design
The backend conversational system included:
➡️ 8 intents
➡️ 4 slot types
➡️ Multiple branching emotional flows
➡️ Retry handling
➡️ Fallback behavior
Intents
Slot Types
This structure allowed Milo to:
➡️ Identify emotional states
➡️ Route users into relevant pathways
➡️ Capture responses
➡️ Adjust future prompts
➡️ Gracefully recover from conversational breakdowns
One of the biggest challenges was balancing flexibility with simplicity. We wanted the system to feel adaptive without creating confusing branching complexity.
Designing Techniques & Interventions
Each emotional state was paired with lightweight interventions designed to work quickly in real-world situations.
The project focused on helping users regulate themselves in the moment rather than creating long-term wellness programs.
Designing for Low Cognitive Load
One of the most important UX considerations throughout the project was cognitive effort.
When users feel overwhelmed, even simple interfaces can become difficult to navigate.
Because of this prompts were intentionally concise, choices were minimized, interactions followed predictable pacing and conversational steps focused on one action at a time
We intentionally avoided:
➡️ Long explanations
➡️ Too many branching choices
➡️ Information-dense screens
➡️ Excessive interaction steps
This became one of the defining UX principles of the project:
✅ Emotional design also means designing for mental capacity.
Reflection & Feedback Loops
A key part of the experience was allowing users to reflect on whether the intervention helped.
After completing an exercise, Milo gently checked in again:
Did the user feel calmer?
Did they want another exercise?
Did they want to continue?
This reflection stage helped the interaction feel adaptive and emotionally responsive rather than purely instructional.
It also transformed the flow from command-response interaction into supportive conversational guidance.
Technical Implementation
Once the conversational flows were finalized, the experience was implemented using AWS Lex and AWS Lambda.
AWS Lex handled:
Intent recognition
Slot capture
Conversation routing
AWS Lambda handled:
Dynamic response logic
Conversational branching
Retry behavior
Adaptive responses
This allowed the system to function as an interactive conversational prototype rather than only a scripted concept.
The technical implementation also reinforced the importance of designing conversations structurally, not just visually.
Mock Conversation Scripts
This project also consisted of creating 10 more mock conversational scripts. Each script covers a distinct scenario: Pre-meeting anxiety, Afternoon crash, Overwhelm spiral, Can't focus, Emotional exhaustion, Restlessness, Calm maintenance, Ambiguous feelings, user wanting to stop mid-session, and a scope-boundary situation (user asking for therapy-level support). Every script includes the emotional state, intent triggered, technique used, and a "Why This Decision" note explaining the design rationale.
Wizard of Oz Testing
To test Milo's conversational design before full technical implementation, we conducted Wizard of Oz (WoZ) sessions, a research method where a human simulates the system's responses in real-time while a participant believes they are interacting with the actual product.
Key Pattern Across WoZ Sessions
Both sessions confirmed the same underlying principle: language that feels neutral or efficient to a designer can feel clinical or demanding to someone in an emotional state. The most impactful revisions from WoZ testing were not structural, they were single word and phrase changes that shifted the emotional register of entire interactions.
Key Challenges
Balancing empathy with clarity
One of the biggest challenges was ensuring Milo sounded emotionally supportive without becoming:
Overly sentimental
Artificial
Repetitive
Emotionally performative
➡️ We learned that emotional design in conversation is often about restraint.
➡️ Too little empathy made the interaction feel robotic. Too much made it feel unnatural.
➡️ Finding the right balance became a major part of the iterative scripting process.
Preventing Conversational Fatigue
Another challenge was avoiding overly long interactions.
During moments of stress, users may not have the energy for extended dialogue.
This forced us to think carefully about pacing, conversational length, response timing and interaction density
We learned that conversational UX is not simply about adding dialogue , it’s about understanding when less conversation creates a better experience.
What I Learned
This project fundamentally changed how I think about conversational interfaces.
Before Milo, I primarily thought about conversation as:
➡️ Task completion
➡️ Interaction efficiency
➡️ Navigation replacement
But Milo showed how conversation can also:
➡️ Reduce emotional friction
➡️ Create reassurance
➡️ Guide emotional pacing
➡️ Support users during vulnerable moments
One of the biggest takeaways was realizing how deeply emotional tone affects user trust. Small wording changes completely shifted whether the interaction felt supportive or transactional.
The project also reinforced that conversational systems should not only be functional, they should account for emotional context, cognitive load, and human vulnerability.
Final Reflection
Milo explored what conversational UX could look like when designed around emotional support instead of efficiency alone.
Rather than trying to maximize engagement or simulate therapy, the project focused on helping users feel slightly more grounded through lightweight conversational interactions.
Ultimately, the project became less about building a chatbot and more about designing emotionally considerate systems for moments when people have the least emotional energy to spare.