HairLinkSG: Know what to ask before salon day
Inspiration
Finding a hairstylist often starts with fragmented social posts and ends with important decisions being made only after the customer reaches the salon. The uncertainty is even larger for newcomers, international residents, and anyone trying a service for the first time. Price, maintenance, expected results, and model requirements are rarely presented in one consistent place.
The original idea came from our Singapore-based teammate, who also led UX planning and the go-to-market strategy for Singapore. Together, we wanted to build an experience where customers and stylists can understand each other before a booking. HairLinkSG combines transparent stylist and model discovery with a structured consultation that turns an unclear concern into something several professionals can answer consistently.
In Singapore's dynamic and diverse market, communication barriers or differing styling expectations often lead to salon anxiety. HairLinkSG tackles this by transforming vague customer requests into precise, structured GPT-5.6 briefs, effectively bridging the communication gap between clients and stylists.
What it does
HairLinkSG is a native SwiftUI marketplace with five connected areas:
- Explore active stylists and model opportunities.
- Quick Match urgent and same-day posts with service and area filters.
- Consult multiple stylists through one structured request.
- Inbox to continue a proposal into a booking conversation.
- Profile to switch between customer/model and stylist perspectives.
The Build Week feature uses GPT-5.6 Terra through Codex. A user enters a concern, desired result, maintenance preference, budget, and area, with an optional photo. GPT-5.6 returns exactly three style directions, cautious observations, three questions for a stylist, and three practical service notes. The output becomes a human-readable brief; it is not a diagnosis or a replacement for professional judgment.
How we built it
Our Singapore-based teammate originated the concept, defined the target personas, structured the consultation flow to reduce psychological barriers, and designed the early-user acquisition strategy. SangHyuk led the technical development, translating that product and UX direction into a working native SwiftUI prototype.
SangHyuk used Codex with GPT-5.6 Sol as an engineering collaborator throughout implementation. Codex helped narrow the technical MVP, implement reusable views and data models, design the structured output contract, add responsible image constraints and error states, build the project repeatedly, and test the complete flow on an iPhone simulator.
The development model and the in-app runtime model have separate roles. GPT-5.6 Sol powered the Codex engineering workflow. For the live feature, the simulator sends the consultation only to a Mac-local bridge bound to 127.0.0.1. The bridge starts an ephemeral, read-only Codex session using gpt-5.6-terra, validates the response against a strict JSON schema, then returns the brief to SwiftUI. The app displays both the runtime model name and the real Codex session ID.
The demo does not require or store an OpenAI Platform API key. It uses the participant's existing ChatGPT-authenticated Codex session and promotional Codex credits. Any optional photo exists only in a temporary file and is deleted immediately after the Codex invocation.
Challenges
The first challenge was scope. A real two-sided marketplace needs authentication, databases, moderation, notifications, and operational support. For one Build Week, we focused on the smallest complete story: discover, describe, analyse, compare, and continue to chat.
The second challenge was using images responsibly. A hair photo can add useful context, but the model must not identify a person, infer sensitive traits, or present cosmetic observations as medical facts. We defined those safety boundaries, and SangHyuk implemented explicit prompt constraints, a schema that requires cautious wording, and an in-person verification reminder in the result UI.
The third challenge was connecting a native iOS demo to the hackathon's Codex credits without pretending that those credits were OpenAI API balance. The solution was a loopback-only bridge that invokes the current Codex runtime directly. That eliminated embedded secrets and produced auditable model/session evidence while remaining honest about the architecture's demo-only boundary.
What we learned
We learned that AI is most useful here when it improves communication rather than pretending to replace an expert. The best role for GPT-5.6 was not choosing a hairstyle with certainty. It was transforming incomplete customer language into consistent goals, questions, and constraints that make the human consultation more productive.
We also learned that product quality depends on the path around the model: input design, privacy boundaries, predictable parsing, loading and error states, visible verification, and a clear handoff to a professional.
What's next
After Build Week, the consumer marketplace direction is intentionally non-AI. The next version will focus on persistent profiles, transparent opportunities, consent, and communication between Singapore customers and stylists. The Build Week experiment remains a working demonstration of how Codex and GPT-5.6 can reduce uncertainty before salon day.
Log in or sign up for Devpost to join the conversation.