Comfort AI

Work conducted for an Automotive Seat Foam Supplier

Introduction & Objective

My role: Pitched Project Idea & Scope

Comfort AI was created to solve a critical knowledge‑retention problem: what happens to decades of institutional expertise when a Subject Matter Expert retires?

The client’s SharePoint library contained thousands of technical documents, but without context or guidance, new engineers struggled to find the right information or know which file to trust.

Project Objective:
Design and deploy an AI agent that preserves expert knowledge, makes documentation searchable and understandable, and empowers engineers to get accurate, source‑cited answers in seconds.

Key Goals:

Transform a static SharePoint archive into an intelligent, searchable knowledge system
Ensure answers are grounded in validated documentation
Maintain strict access controls and data security
Build trust through transparency, UX clarity, and verifiable citations

User Research

My role: User Research Lead

Stakeholder Interviews

Spoke with engineers, quality specialists, and the retiring SME
Identified pain points around search, terminology, and document interpretation
Mapped how experts historically navigated the knowledge base

Document Ecosystem Audit

Reviewed thousands of SharePoint files
Categorized content types (terminology, SMTE docs, industry references, technical papers, etc.)
Identified inconsistencies, gaps, and patterns in how information was stored

User Workflow Analysis

Observed how engineers currently searched for answers
Documented friction points: slow retrieval, uncertainty, lack of context, inconsistent naming conventions

AI Research

My role: AI Research Lead

AI Feasibility & Constraints

Evaluated SharePoint integration options
Assessed security requirements, access permissions, and data‑handling rules
Identified risks: hallucinations, outdated files, ambiguous queries

Ideation & Concept Design

My role: Lead Designer

Experience Principles

Trust through transparency (always cite sources)
Clarity over complexity (simple, predictable interactions)
Security by design (respect SharePoint permissions)
Scalability (easy to update as new documents are added)

Information Architecture

Created a structured knowledge map of the SharePoint library
Defined how the agent should prioritize sources
Designed logic for handling ambiguous or multi‑file answers

Interaction Design

Drafted conversation flows
Designed fallback behaviors and fail‑safes
Created UI concepts for side‑panel vs. full‑screen modes

Prototypes & Evaluation

My role: Lead Prototyper & Lead Tester

I built a functional prototype of the AI agent and tested it with real engineering questions.

Prototype Features

Natural‑language question handling
Source‑cited answers
Document previews
File‑level access control
Ability to learn from new uploads

Usability Testing

Conducted sessions with engineers and commercial team members from the client's company across experience levels.

What I tested:

Accuracy and clarity of answers
Trust in citations
Ease of navigating between files and responses
Confidence in using AI for technical decisions

Key Findings:

Users trusted answers significantly more when citations were visible
Engineers preferred concise summaries with optional deep‑dives
The side‑panel interface reduced context‑switching
Fail‑safes (e.g., “I don’t have enough information”) increased credibility

Iteration
Based on feedback, I refined:

Prompt structure
Error‑handling logic
UI layout
Document prioritization rules
Language tone for explanations

Final Product

My role: Product Owner & Developer

The final Comfort AI agent is a fully integrated SharePoint‑embedded system that:

Core Capabilities

Answers engineering questions related to seat comfort using validated documentation
Cites exact files and sections used
Adapts to new uploads and updated documents
Respects SharePoint permissions and access levels
Works in both side‑panel and full‑screen modes

UX Features

Clean, minimal interface
Clear citation formatting
Expandable document previews
Smart fallback responses
Quick Links for fast access

Technical Highlights

Custom prompt engineering
Knowledge mapping and data structuring
SharePoint integration with secure access controls
Scalable architecture for future document sets

Impact & Reflection

Impact

Faster onboarding: New engineers find answers in seconds instead of hours
Knowledge preservation: SME expertise remains accessible long after retirement
Higher confidence: Source‑cited answers reduce uncertainty and rework
Improved consistency: Terminology and decision‑making become more aligned
Scalable framework: The same architecture can be applied across departments

Reflection

This project reinforced the importance of designing AI systems that are:
Human‑centered
Transparent
Secure
Context‑aware

The most rewarding part of this project was watching users shift from skepticism to trust, not because the AI was flashy, but because it was reliable, verifiable, and intentionally designed around their real workflows.

I also gained a deeper understanding of how AI integrates into existing systems and how people actually interact with it in high‑stakes environments. Seeing where AI excels, where it struggles, and how those limitations must be accounted for was invaluable. It reinforced that building an effective AI agent isn’t just about accuracy, it’s about creating a tool that improves efficiency, earns user trust, and provides answers that can be fact‑checked and relied on.

This project was very enjoyable, and it was great to launch this project to not only the engineering team but to the key stakeholders in the company as well