A static portfolio could not explain enough project context, target-role fit, or AI workflow experimentation without becoming too dense.
02
What I built
Created Ask Leo as a recruiter-facing assistant backed by curated profile and project knowledge, with rate limits and cost controls kept server-side.
03
Outcome
Ask Leo becomes a differentiator: an interactive, evidence-aware career lab rather than a generic chatbot.
Screenshots
See it for yourself.
click any image to zoom
Shipped
✓AI portfolio assistant with curated knowledge sources
✓Public resume, blog, and project evidence paths
✓Private owner tools separated behind login
Hard parts
!Keeping answers concise and recruiter-friendly.
!Using AI without exposing private workspace data or uncontrolled costs.
!Separating public showcase modules from owner-only daily tools.
Build Story / Interactive case file
How I turned my portfolio into a system I actually use.
I built MyJournal in visible repository checkpoints: first the private daily and resume tools, then Ask Leo and its guardrails, then a stronger public evidence layer. This is the development path recorded in the project history.
01
Foundation
May 18, 2026 · 5e18375
I started by making the site useful to me, not only presentable.
My first major step was a local-first Personal Dashboard and Resume Assistant: one place for daily reflection, another for preparing job-specific material.
That checkpoint changed MyJournal from a portfolio I maintained for visitors into software I could use myself. I kept public pages intact while adding private daily records, local backup, AI companion support, and editable resume drafts.
May 18 workspace milestone
Repository checkpoint 5e18375: Build local-first personal dashboard and resume assistant.
03
AI lesson
June 7–8, 2026 · 70fc022 / b720a10
I added Ask Leo, then tightened it one day later.
The first Ask Leo release made the portfolio conversational. The next iteration addressed the practical problems I saw: token cost, irrelevant context, privacy, and unsupported answers.
I changed it from a broad portfolio chatbot into a routed public assistant. Questions are classified, private requests follow a separate path, and the model receives a small relevant evidence set instead of the whole knowledge base.
Send the whole portfolio
Not chosen
Simple prompt design, but high token use and weak relevance.
Mix public and private context
Not chosen
More context, but an unacceptable privacy boundary.
Classify, retrieve, then answer
Selected
Small evidence set, class-specific limits, public information only.
Checkpoint b720a10 records the follow-up optimization: classification, retrieval trimming, caching, rate limits, and class-specific output budgets.
05
Evidence layer
June 22–July 28, 2026 · c4770f3 / 5834603 / 4168534
I rebuilt the public side around inspectable project evidence.
I added dedicated project routes, case-study interactions, LeoOS Lab, and later Project Universe. Build Story is the next step: explaining decisions and iterations instead of showing only final screens.
The result is still one evolving personal project, but with deliberate boundaries: public evidence and Ask Leo for visitors; Daily Center, resume preparation, finance, and Job Center for owner workflows.
Public user path
Recruiter-facing Q&A
Ask project and role-fit questions from the public portfolio.
Implemented
Server-side AI controls
Classification, source selection, rate limits, caching, and budgets.
Boundary
Public / private separation
Daily records and owner workflows are not used as public assistant context.
Repository evidence
Repository-backed story
Screenshots and code excerpts point to material that can be inspected.
This story reports implemented behavior and product boundaries. It does not claim automatic job applications or unrestricted access to private data.
May 18 workspace milestone