Conversation-derived project canvas
Prototype Review
Workspeed / Luxava voice AI call assistant
Visual notebook extracted from one free-flowing founder conversation about a reactive voice agent for med spas and beyond.

Project Summary

  • Project name: Workspeed (prototype engine) / Luxava (product brand direction)
  • What is this? A voice-first AI assistant that joins real calls and responds only on cue, helping people pull in information or assistance “on their terms”.
  • Type: Productized AI service / SaaS-like call companion
  • Primary audience: Initially med spas & aesthetic clinics; conceptually usable by anyone on calls.
  • Primary value: Let users quickly query or instruct an AI mid-conversation without constant interruptions.
  • Current stage: Early prototype + naming/positioning exploration; debugging reactivity and wake word handling.
Voice AI
Reactive “wake-word” mode
Med spa GTM
Zero-shot HTML summarizer
Snapshot
The founders are testing a live Workspeed prototype, critiquing its “reactive mode” behavior, admiring an automatic HTML summary output (Luxava app), and aligning on market scope, naming, and next steps.
  • Reactive mode only triggers when user begins a sentence with “Workspeed”.
  • This causes missed requests inside natural speech → identified as a key bug.
  • Luxava HTML visualization is generated from prior call and reviewed here.

Conversation Timeline

From first hello to next steps
Friend • Prototype deployment

Uploads Workspeed mobile file & shares link

Friend has just generated a mobile build and is hosting it on their own domain for quick testing.
Friend • Market probe

Med spa market question as test case

Uses “How many med spas in the US?” as a representative in-call query, illustrating how often people want quick facts mid-conversation.
You • Interaction model

Explains active vs reactive vs passive modes

Clarifies that the limited “reactive” behavior is intentional: users must start sentences with “Workspeed” so the agent doesn’t interrupt organically.
Friend • UX critique

Notes segmentation bug with wake word

Repeated “Workspeed…” queries fail; you infer it only listens when the entire sentence starts with “Workspeed”, ignoring mid-sentence triggers.
You & Friend • Debugging

Root cause: sentence segmentation logic

You deduce the simple fix: detect the keyword “Workspeed” anywhere in the utterance, then capture the tail of that utterance as the prompt.
Both • Meta review

Review Luxava HTML output from previous call

You open call.luxava.app, admire the visual excitement/risk tracks and system swimlane, and recognize it’s essentially a zero-shot summarization that can be further improved.
Friend • Emotional reflection

Aligns on productivity vs outcome

Friend felt last week was less productive than hoped, but now feels stoked seeing the concrete outcome and progress on the product.
You • Market & focus

Growing excitement & hesitancy about focus

You admit earlier hesitation due to focus risk, but the broader applicability and low barrier to entry feels increasingly exciting.
Friend • Naming & ownership

Discuss Luxava naming and Workspeed ownership

Friend will generate naming options like “Luxava”; you confirm Workspeed is owned personally (via GoDaddy) and not encumbered by prior collaborator Chris.
Both • Close-out

Plan rest + follow-up work

You agree to rest, regroup later that day, and push forward on bug fixes, naming exploration, and go-to-market planning.
Emotion / Alignment Track
🤷
✨
⚠️
✨
⚠️
✅

Stakeholders & Roles

Who cares & why
Founders / Builders
Vision + Execution
Goals
  • Ship a reliable, intuitive voice agent that behaves on users’ terms.
  • Validate med spa / aesthetic clinic market while keeping product broadly useful.
  • Develop a clear brand (Workspeed vs Luxava) and a scalable narrative.
Pain points
  • Wake word segmentation bug causing frustration in live tests.
  • Tension between deep focus and huge horizontal opportunity.
  • Perception that time spent last week wasn’t maximally productive.
What this project does
Provides a concrete product to rally around, with visible progress (HTML summaries, live prototype) that increases motivation and clarity.
Med Spas & Aesthetic Clinics
Initial go-to-market segment
Goals
  • Handle client calls efficiently while offering knowledgeable responses.
  • Reduce manual effort on repetitive information lookup during calls.
Pain points
  • Need to quickly reference data (pricing, services, availability, market info) mid-call.
  • Limited staff attention and time; interruptions are costly.
What this project does
Lets staff invoke “Workspeed” mid-call to get instant assistance or answers without leaving the conversation or juggling multiple tools.
General Call Participants
Broader horizontal market
Goals
  • Seamlessly “talk to an AI” during meetings, sales calls, or personal conversations.
  • Access real-time information or AI assistance verbally.
Pain points
  • Frustration when AI assistants randomly interject instead of responding on cue.
  • Current assistants often locked to devices or require explicit UI interaction.
What this project does
Offers a call-embedded, voice-activated agent that respects conversational flow and responds when explicitly addressed.
Prior Collaborator “Chris”
Contractual / IP considerations
Goals / Concerns
  • Clear boundaries regarding ownership of Workspeed assets.
  • Formal release of any potential claims in new contract.
Impact of this project
The founders intend to ensure Workspeed is owned cleanly, with contractual language releasing Chris from future claims and vice versa.

Project Canvas

Domain-neutral summary
Problem / Opportunity
People frequently need to reference facts or get AI support while already talking to someone. Existing assistants either:
  • Interrupt conversations unpredictably, or
  • Require device-centric interaction that breaks flow.
Opportunity: a reactive, in-call AI that responds only when explicitly invoked, starting with med spa call workflows but expandable to nearly any call context.
Target Audience
  • Primary: Owners/staff of med spas & aesthetic clinics conducting many phone calls.
  • Secondary: Knowledge workers, sales teams, and everyday users wanting a voice AI “sidekick” during conversations.
The design is intentionally broad: “literally anybody could use it”.
Value Proposition
  • Ask the AI anything in the middle of a call using a wake word, without the agent randomly chiming in.
  • Maintain conversational flow while accessing real-time info or assistance.
  • Automatically generate rich visual summaries (like the Luxava HTML canvas) from calls for later review.
Key Activities / Capabilities
  • Wake-word detection tied to audio segmentation (reactive mode).
  • Mode switching between active, reactive, and passive behaviors.
  • Audio capture, transcription, and prompt extraction for the AI model.
  • Generation of structured visual summaries of conversations (HTML canvases).
  • Naming & brand development (e.g., Luxava) and keyword research for GTM.
Constraints / Limitations
  • Current prototype only recognizes “Workspeed” at the very start of an utterance.
  • No direct access to real-time external data (e.g., live weather, precise med spa counts) is evident in this test.
  • Low barrier to entry means potential competition and commoditization risk.
  • Founder time and focus are limited; need to avoid scattering across too many directions.
Success Metrics / “Good” Looks Like
  • Wake word reliably recognized even in the middle of natural speech.
  • Users feel in control: assistant never “randomly chimes in”.
  • Med spa staff find it obviously helpful and adopt it in daily calls.
  • Conversation summaries (like call.luxava.app) are clear, accurate, and delightful.
  • Founders feel their time with the project is highly leveraged and exciting.

Insights, Decisions, Risks & Next Steps

Linked back to quotes

Decisions

  • Reactive mode is the default for founders Active for people who want constant guidance (e.g., your son); reactive for you; passive for quiet observation.
  • Fix wake-word segmentation logic Change detection to trigger on “Workspeed” anywhere in the utterance and capture the following words.
  • Use med spas as an initial GTM beachhead Broader market potential recognized, but med spas remain a concrete starting narrative.
  • Chris will not own Workspeed You confirm Workspeed is personally owned and contract will release him of claims.

Key Insights

  • Mid-conversation lookup is the killer use case The med spa count example illustrates a frequent desire to ask the AI “real quick” during live calls.
  • Zero-shot HTML summarizer already feels impressive The Luxava output “looks pretty dope” even with almost no prompt engineering.
  • Perceived productivity is offset by visible outcomes Even if last week felt less productive, tangible artifacts reframe the effort positively.
  • Huge horizontal scope is both exciting and risky “Literally anybody could use it” → massive opportunity but raises focus concerns.

Non-goals / Rejected Ideas

  • No always-on, interrupting assistant Random chiming in is considered unacceptable; the wake-word constraint is deliberate.
  • Not married to “Workspeed” as external brand Open to rebranding (e.g., Luxava) especially as scope broadens.

Risks / Concerns

  • Wake word unreliability undermines trust Current behavior feels “hosed” and fights with the user; must be fixed quickly.
  • Focus dilution due to broad applicability You worry about spreading effort too thin despite the enticing market size.
  • Low barrier to entry invites competition Easy entry makes differentiation via UX, brand, and ecosystem more important.

Open Questions

  • How will Workspeed securely access real-time external data? Weather / med spa count prompts point at data-source and reliability questions.
  • Is med spas niche or just the first marketing wedge? How narrow should early positioning be vs. horizontal expansion?
  • What final brand architecture: Workspeed vs Luxava? Engine vs app vs company naming is not yet resolved.

Next Steps

  • Implement the segmentation fix for wake word Change logic to trigger whenever “Workspeed” appears, then parse the rest of the utterance.
  • Generate name & keyword exploration report Friend will run the naming script (which yielded “Luxava”) and prepare options for review.
  • Rest, then reconvene to push bug fixes and GTM planning Coordinate after attorney meeting / rest to continue momentum.

Trade-offs & Alternatives

Design choices in the conversation
Interaction Modes
Mode Pros Cons
Reactive (wake-word only)
  • Respects conversation flow; no random chiming in.
  • Ideal for power users who only want help on demand.
  • Requires robust wake-word detection; current segmentation bug hurts UX.
  • Extra cognitive load: users must remember to say “Workspeed”.
Active (always responding)
  • Great for users needing constant guidance (e.g., your son or less technical friends).
  • Feels like a continuous companion or tutor.
  • Risk of “randomly chiming in” and hijacking conversation.
  • Not suitable when users mention the product name casually.
Passive (silent observer)
  • Safe in contexts where hearing its own name would be disruptive.
  • Can still record / summarize without intervening.
  • No in-the-moment help; only useful for after-the-fact artifacts.
Brand Scope: Workspeed vs Luxava
Option Pros Cons
Luxava as outward brand
  • Fresh identity discovered via naming script.
  • Feels aligned with broader, possibly consumer-friendly market.
  • Not tied to prior collaborator relationships.
  • Less attached to current prototype name; may create mental mapping overhead.
  • Needs validation that it resonates with med spa market.
Workspeed as primary brand
  • Already owned personally and technically in use.
  • Evokes “faster work”, aligning with productivity angle.
  • Friend is “not married to it” for a broad, possibly more lifestyle-oriented product.
  • Historical linkage with Chris; though legally clean, may carry baggage.

Topic Map & Idea Flow

How the conversation wandered
Wake word & reactive mode “You have to lead with the words Workspeed…”
Med spa market example “We’re discussing a product… med spas and aesthetic clinics”
HTML call canvas (Luxava) “See that little excitement… system view”
Debugging segmentation “All you have to do to fix that is…”
Productivity & emotions “Time last week… look at this cool little outcome”
Focus vs broad market “Possibilities are endless… barrier of entry”
Naming & ownership “That is how we ended up landing on Luxava… GoDaddy”
Narrative Flow
1
Launch & test Workspeed mobile build in a real environment (coffee shop).
2
Use med spa count question to stress-test wake-word and in-call experience.
3
Notice failure cases in reactive mode → diagnose segmentation bug → identify fix.
4
Open Luxava HTML summary from previous call; evaluate visualization quality and possibilities.
5
Reflect on productivity, excitement, and focus risk now that outcomes are visible.
6
Discuss broader market scope, naming (Luxava vs Workspeed), and IP ownership.
7
Plan rest, legal follow-up, keyword/naming work, and deeper product iteration.

System & Process Views

How it operates at a glance
High-level Call Flow
Caller & Staff
Have a live conversation (e.g., client asking about med spa services).
➜
Workspeed Listener
Listens to audio stream in reactive mode, waiting for “Workspeed”.
Wake Word Detected
User says “Workspeed, how many med spas are there…”
➜
Segmentation & Prompting
Extracts everything after “Workspeed” as the AI prompt.
AI Engine
Processes query and returns an answer based on available knowledge / tools.
➜
Voice Output
Speaks response back into call, then returns to listening mode.
Post-call Summarization
Recorded Conversation
Audio + transcription of the session.
➜
Summarization Engine
Converts raw transcript into structured insights: timeline, stakeholders, risks, etc.
Luxava HTML Renderer
Renders a visual canvas like call.luxava.app with emotion tracks and system views.
➜
Founders & Users
Review, refine, and share the summary; use it as a thinking/decision artifact.
Conceptual Entities
  • Call Session: id, participants, start/end time, audio, transcript, mode (active / reactive / passive).
  • Invocation: timestamp, raw utterance, detected wake word position, extracted prompt, AI response.
  • Summary Artifact: call_id, sections (timeline, canvas, risks, decisions), emotion track, trade-offs.
  • Brand Concept: name (Workspeed, Luxava), domain, narrative, associated market segment.
Relationships
  • One Call Session has many Invocations.
  • Each Summary Artifact belongs to one Call Session.
  • Multiple Call Sessions can be associated with the same Brand Concept and target segment (e.g., med spas).
Key Sequences
sequenceDiagram participant User as Caller / Staff participant Call as Call Audio participant WS as Workspeed Listener participant Seg as Segmentation Logic participant AI as AI Engine participant Summ as Summarizer (Luxava) User->>Call: Talks with other person Call-->>WS: Streaming audio User->>Call: "Workspeed, how many med spas..." Call-->>WS: Utterance containing wake word WS->>Seg: Detect "Workspeed" and cut prompt Seg->>AI: "How many med spas are there in the US?" AI-->>WS: Response text WS-->>Call: Speaks answer back into call Note over WS,Seg: Current bug: only triggers if utterance starts with "Workspeed" Call-->>Summ: Full transcript after call ends Summ-->>User: Visual HTML canvas (timeline, risks, etc.)

Transcript Traceability

Click insights to see source quotes
Every decision, risk, or topic node is tagged with a quote id. Click any bullet above to highlight the related quote here or open it in a modal.
Q1 • Wake-word instruction
Friend: “So I think you have to lead with the words Workspeed. You can't say, hey, work speed.”
Q2 • Med spa example
Friend: “Hey, Workspeed, we're discussing a product that we're going to sell to med spas and aesthetic clinics. How many med spas are there in the United States? Roughly?”
Q4 • Intentional mode design
You: “That's by design, by the way. That's not an accident. That feature's in there like that so that you can ask it on your terms and not have it, you know, just fucking randomly chiming in all the time. The other mode, the active mode, is really fucking good for people like my son… and then the reactive mode is good for us, and then the passive mode is good for particular scenarios…”
Q6 • Reaction to Luxava HTML
Friend: “Jesus, dude. Wow. See that little excitement, concern, risk, ambiguity, alignment things. Yeah, that's pretty dope… what it spits out is pretty impressive… I like that swim lane at the bottom with the system view.”
Q8 • Friction with prototype
Friend: “Work speed, set your mode to active. Damn thing's just fighting with me… Okay, whatever. It's hosed now.”
Q9 • Segmentation bug & fix
You: “The segmentation of the way that we're speaking, it only looks for the beginning of a sentence that begins with Workspeed… all you have to do to fix that is just change it to see if it detects anything that says the word Workspeed, and then the words after that until the user stops speaking is what you want to use for addressing it. I think that's the glitch, right?”
Q13 • Plan to reconvene
You: “Anyways. Yeah, chew on that. I'm gonna get like a hour happen or so and wake up in a bit here… I'll give you a shout when I get done with this attorney or I'll just wait to hear from you later this afternoon.”
Q14 • Med spas as product focus
Friend: “We can do this in a few ways… we're discussing a product that we're going to sell to med spas and aesthetic clinics…”
Q15 • Productivity reflection
Friend: “Hey, time last week did not as productive, but look at this cool little outcome of it.”
You: “Oh, dude, I thought it was great… I have a high standard for myself and so I just had hoped to accomplish a lot more. But I also had a really good time.”
Q16 • Excitement vs focus risk
You: “I was feeling like a little, you know, hesitant at first… I always get worried about the risk of focus. But this thing is… a much broader market. Possibilities are endless… It's just much easier barrier of entry too, which is inherently a bit of a problem, but it's also incredibly enticing.”
Friend: “Well, and the application is. I mean, literally anybody could use it.”
Q17 • Ownership clarity
Friend: “What does Chris own any part of work speed?”
You: “No, it's in my… it's owned in my GoDaddy… there's no issue there and our contract is gonna release him of any claims.”
Q18 • Openness on naming
Friend: “If you don't want to with words be. I don't know. I'm not. I'm not married to it…”
You: “I mean if it's going to be as broad as it looks it. I think it fits perfectly the way that I'm thinking about it currently but yeah, we can ideate on that.”
Q19 • Luxava naming script
Friend: “I can start, you know, start putting together some keyword ideas… I built that script that generated all those things. That is how we ended up landing on Luxava.”
Q20 • Next actions & rest
Friend: “Yeah, go, go take care of yourself and we'll touch bases this afternoon… I'll get a plug in plugged into our script… then I'll put a report out in front of us. We can poke through to figure out, you know, brands and stuff.”

Related quote