Work Tool · AI
Guided Self-Review Assistant
Screenshot shows sanitized example data, not real customer or employee information.
A lightweight guided web app that walks employees through structured mid-year self-reflection questions one at a time, then uses an LLM to draft polished, evidence-based answers covering accomplishments, growth areas, and feedback for leadership. Built as a dependency-free vanilla JS front end with an iterative refinement loop and a regional variant for distributed teams.
Architecture
A dependency-free front end that walks someone through six fixed questions, then drafts their self-review using their own pre-loaded accomplishment notes as grounding for an LLM.
Step 1
Personal accomplishment log
Each person's own notes are pre-loaded and mapped to company competency principles ahead of time.
Step 2
Guided question stepper
A vanilla JS stepper enforces one question at a time, so the flow can't be skipped or rushed.
Step 3
Single structured draft call
Once all answers are collected, one structured prompt asks the LLM to draft three sections: accomplishments, growth areas, and leadership feedback.
Step 4
Iterative tightening
A follow-up conversational loop lets the person ask for edits without losing the original context.
Key decisions
- Enforced the question order in the UI rather than trusting the model to ask one thing at a time, making the experience far more predictable.
- Kept the accomplishment data grounded in the person's own notes only, explicitly preventing the model from inventing achievements or borrowing someone else's.
- Shipped a second regional variant by swapping only the roster/data file, keeping the app logic identical.