Ask Nate

[Independent · 2026]

A conversational portfolio experience. Visitors ask about my work, get evidence-backed answers, follow approved case study routes, and see ASCII motion while replies prepare.

Portrait of Nate Johnson against a black background

Product brief

Ask Nate is a conversational portfolio experience that helps visitors explore my work, experience, and design approach through direct questions and relevant case study recommendations.

Visitors ask about projects, roles, skills, and experience, then receive a concise answer and relevant case study recommendations. The product needed to feel useful without asking visitors to trust fluent but unsupported prose.

Early studies tested the character, ASCII portrait, expression language, scale, and question to follow up journey before the product direction settled.

The solution

01 / 05 1

ASCII reply lifecycle

Ask Nate uses thinking, searching, generating, and complete as the primary reply sequence, with a distinct error state when the request cannot finish. The same ASCII engine supports empty states, login, and the missing page experience. Reduced motion holds a static frame while preserving the announced state.

02 / 05 2

A structured evidence model

Fluent answers are not the same as accurate answers, so every reply is grounded in seven sources of truth. Base knowledge is always present; when a question names a live project or the site itself, that topic’s full published evidence joins the prompt deterministically. The prompt requires direct answers, honest partial answers instead of dead ends, and recommendations only when relevant.

Ask Nate visual direction with green character shapes and introductory portfolio copy

03 / 05 3

Clear ownership between model and interface

The model should not own navigation. It understands the question, selects evidence, writes the response, and suggests a next step. The interface controls links, project components, feedback states, ASCII animation, loading, and errors. Visitor question → prompt rules and structured evidence → model response → application controlled routes and presentation.

Ask Nate experience map showing preset and input paths into animation, result, and follow up

04 / 05 4

A repeatable evaluation workflow

A 57 question regression matrix runs after prompt or knowledge changes, covering every live project, employer chapters, aliases and misspellings, contextual follow ups, partial evidence, and refusal protection. Each question has an expected evidence source and next action, and automated tests keep the knowledge set, live routes, and public project registry aligned.

Ask Nate ASCII expression set beside a large portrait study

05 / 05 5

Application controlled project links

The application removes generated link lines, matches project titles and approved aliases against the live project registry, and attaches known routes after the response is written. Hidden projects remain unlinked, and tests assert that the registry contains only live projects.

Ask Nate introduction paired with an ASCII portrait and green status character

Grounded answers.
Approved routes.
Visible waiting.

By the numbers

Ask Nate character scale study with rounded rectangle and circular expressions

repeatable questions in the regression matrix

structured sources of portfolio truth

primary reply states before error handling

My role

I designed and built the conversational prototype, prompt rules, structured evidence model, project registry, link attachment behavior, chat interface, and ASCII feedback language. I also created the repeatable question set and automated checks that keep knowledge, live routes, and starter questions aligned with the published portfolio.

Role

Product Designer

Timeframe

2026

Disciplines

Product Design, Interaction Design, Design Systems

Current state

Ask Nate ships with structured evidence, approved routes, the 57 question regression matrix, automated knowledge and link checks, randomized starter questions, and a shared ASCII feedback language. The live chat runs on this site: and ask about this case study.

Evidence model

Structured project fields and evidence categories feed prompt-backed answers

Link attachment

Approved live routes only. Exact title and alias matching. No guessed URLs

Feedback language

ASCII motion with primary states and micro behaviors shipping sitewide, including reduced motion

Platform

Web (Next.js)

Storyboard of every frame in the five ASCII reply states: thinking, searching, generating, complete, error — ASCII reply lifecycle · every frame · thinking, searching, generating, complete, error

The twelve primary ASCII states, first frame and behavior verb for each — ASCII Motion Design System · twelve primary states

Problems & decisions

Help visitors explore my work through direct questions while keeping every answer grounded in verified portfolio evidence and every project link under application control.

A conversational layer could make a large portfolio easier to explore, but invented claims, inaccurate recommendations, or unpublished links would weaken trust in the work it was meant to explain.

Visitors needed direct answers to specific questions, a useful next step, and clear signals while a reply was preparing. They should not have to verify whether a recommended project or route was real.

Should the model generate the complete experience, including project links, or should the application control evidence, routes, and presentation? I let the model understand the question and write a concise response from supplied evidence, while the application attaches approved routes, manages interface states, and presents project components. The tradeoff: factual claims still depend on the model following the evidence rules, so deterministic claim validation remains an open improvement.

Core response rules

Use only the provided portfolio evidence.
Do not invent project details or experience.
State when information is unavailable.
Answer the question directly.
Recommend projects only when relevant.
Never create or guess portfolio links.

Sources of truth

Project data
Structured project evidence
Published case study content
Portfolio design system reference
Approved project routes

Question	Expected evidence	Next action
Which projects best show your product design work?	Role fit product set with at most five live titles	Live case study links
What design systems have you built?	The design system behind this site, and the engines shipped as libraries	Live design system case studies
How does the ASCII engine work?	Engine case study evidence only	Live engine case study
What results can you verify?	Project bound metrics only	Named result projects
Do you have a resume?	Nothing in the knowledge block	Decline and offer email
Where have you worked?	Nothing in the knowledge block	Decline and offer email

Ask Nate only stays accurate when evidence and routes are application-controlled. Prompt rules ground the voice, but claim validation before display is still an open improvement. Waiting works when feedback is a shared ASCII language, not a one-off loader. Deterministic claim validation and a recorded chat walkthrough remain open improvements.