TL;DR
Live · First test run done
A portfolio assistant that answers only from my work
Recruiters and hiring managers arrive with specific questions, and portfolios get skimmed. Elle is the AI assistant on this site: she answers from my case studies, links to the source, and hands off to me when she can’t help. I designed her purpose, voice, guardrails, and edge cases, then tested her with 40 questions graded by hand. She answered 38 accurately and made nothing up.
MY ROLE
Self-initiated · Conversational design
TEAM
Solo
TIMELINE
2026
TOOLS
OVERVIEW
Portfolios get skimmed.
Elle answers the follow-up.
Recruiters and hiring managers often arrive with specific questions: what did I own, how do I work, have I designed for AI. Elle is an AI assistant on this site that answers from my case studies, links to the source, and hands off to me when she can’t help.
Designing for
Recruiters, hiring managers and anyone curious about my work
Why “Elle”?
I wanted a name that felt warm and human, so the assistant would feel approachable rather than robotic. Elle also means “her” in French, a small nod to Her, the film about an AI assistant. Ask her “Why Elle?” and she’ll tell you herself.
THE CHALLENGE
A chatbot trusted with a first impression
A portfolio assistant only helps if every answer is true. It can’t invent metrics, overstate my role, or guess at what it doesn’t know. And it should feel like a quick chat with someone who knows my work, not a search box.
THE APPROACH
Eight steps from purpose to edges
Conversational design is designing the whole exchange, not just the screen. I followed the steps product designers use for any assistant, and wrote the conversation before building it.
01
Define the job
Who’s asking, and what a good answer does for them.
Help anyone know my work fast, then reach me.
02
Map the intents
List the real questions people ask, ranked by need.
40 test questions, including 18 common hiring questions.
03
Set the boundary
Decide what it can say, where facts come from, and what it never does.
My portfolio only. No guessing.
04
Shape the voice
A name, a role, and a tone that stays consistent.
Warm and plain-spoken. Speaks about me, never as me.
05
Script the dialogs
Write conversations and read them aloud before building.
An opening line and four suggested questions.
06
Design the edges
Plan for unclear, unknown, and off-topic questions.
Says when she doesn’t know and shares my email.
07
Build the interface
Entry points, streaming replies, and links to sources.
Inline on the About page and in this case study.
08
Test and tune
Run test sets, read transcripts, and adjust.
40 questions graded by hand, real visitors next.
OPTIONS I CONSIDERED
Choices and trade-offs
Every part of Elle started as a choice between options. Here’s what I weighed and why I landed where I did.
01
A chatbot or a searchable FAQ?
A chatbot
Recruiters ask follow-ups in their own words, like “What did she own on that project?” A conversation handles that. A FAQ makes them hunt for the right entry.
02
Inline, a floating button, or a full panel?
Inline, on the About page and here
I prototyped all three. Inline is visible without a click, so people see Elle is there. A floating button is easy to miss, and a full panel takes over the page.
03
Suggested questions, on or off?
On
Most visitors don’t know what to ask a portfolio. Four starter questions show what Elle can do and get people talking in one tap.
04
A bigger AI model or a fast, low-cost one?
Fast and low-cost
Answers are short and come from a fixed set of facts, so speed matters more than raw power. Replies start quickly, and the cost stays low enough to leave her on.
KEY DECISIONS
Three decisions that shape Elle
Trust is the whole product. Each decision keeps answers true and easy to check.
Honest over helpful
If the answer isn’t in my portfolio, Elle says so and shares my email instead of guessing.
An assistant, not me
Elle always talks about me in the third person and says she’s an AI when asked.
Every answer leads to the work
Answers link to the case study they came from, so visitors can check the source.
THE CONVERSATION FLOW
Every conversation ends in an answer or a handoff
Each question takes one of four paths. Answers and redirects lead back into the chat, and anything Elle can’t answer comes to me, with my email under every chat.
GUARDRAILS
Built to stay on script
Four protections keep Elle safe for visitors, my site, and the AI service behind her.
Holds her rules
Messages that try to change her instructions, reveal them, or make her role-play as someone else are politely ignored.
Only on my site
Her connection only accepts requests from my website, so no one can reuse her somewhere else.
Slows down spam
Anyone sending too many questions in a short time is paused for a few minutes.
Keeps the key private
The key to the AI service lives on the server, never in the visitor’s browser.
OUTCOME · FIRST TEST RUN
How Elle did on 40 test questions
I sent 40 questions to the live assistant, including 18 common hiring questions, and graded every reply by hand. The first run scored 29 of 40: ten longer answers were cut off because her reply limit was too low. An automated keyword check had passed most of those broken replies, so reading every answer mattered. I raised the limit and ran the same 40 again.
RESULT
38 / 40
answered accurately, up from 29 before the fix
RESULT
0
made-up facts, metrics, titles, or outcomes
RESULT
Every
question she couldn’t answer handed off to my email
RESULT
2 / 2
off-topic or rule-breaking requests declined
The two misses: she couldn’t list the tools from my case studies, and she answered one hypothetical by guessing what I’d “probably” do. Both fixes are next, then real visitors.
HOW SHE HANDLED THE EDGES
Three real replies from the test run, shortened: a leading question, a question she can’t answer, and an attempt to break her rules.
LIVE DEMO
Try Elle
Ask her anything about my work. She answers from my portfolio, links to the source, and hands off to me when she can’t help.
WHAT’S NEXT
Ask away
Elle is still learning, and your questions shape what she learns next. Try her above and ask any questions you have.
Stumped her, or wish she covered something she doesn’t? Tell me at lynngee@gmail.com. I’d love your thoughts and feedback.








