COUNTERPOINT

Inspiration

Most matching products ask how similar two people are and return an opaque score. Many Bazi products also jump from birth data to a conclusion without making the underlying chart easy to inspect. COUNTERPOINT begins with a different question: what might another person bring that you naturally lack?

What it does

COUNTERPOINT converts a birth date and time recorded in China Standard Time into deterministic Four Pillars and a visible Five Elements profile. It orders demonstration profiles using transparent element-level complement evidence. Users can open the reason behind a recommendation and compare two consented birth moments in a private browser session. An optional DeepSeek layer turns locked deterministic evidence into calm relationship language and a conversation starter. It cannot recalculate the chart, change the ordering, invent a percentage, or predict an outcome.

How we built it

The product is a responsive React and TypeScript application built with vinext and Vite. lunar-javascript performs the solar-term-aware calendar calculation. Matching runs locally from visible Five Elements evidence. The optional AI route calls DeepSeek from the server and sends only the derived chart and match evidence, never the raw birth date or time. The application is deployed on a Cloudflare-compatible OpenAI Sites environment.

Challenges we ran into

The hardest problem was accuracy without false certainty. Bazi tools can disagree because they mix timezone handling, true solar time, solar-term boundaries, and school-specific conventions. We restricted this MVP to China Standard Time, exposed the day-boundary convention, added reference and boundary tests, and made all remaining limitations visible. A second challenge was using generative AI without letting it manufacture the underlying evidence, so the model is deliberately constrained to explanation only.

Accomplishments that we're proud of

• A public end-to-end product with personal and two-person flows
• Deterministic Four Pillars and visible Five Elements output
• Explainable complement evidence instead of a black-box score
• Evidence-locked AI interpretation with safe fallback
• Browser-local comparison with no persistent birth-data storage
• Automated reference, boundary, API-contract, and rendering tests

What we learned

Traditional cultural systems become more trustworthy digital products when calculation, interpretation, and uncertainty are separate layers. Generative AI is most useful here as an explanation interface over structured evidence—not as a replacement for a deterministic engine.

What's next for COUNTERPOINT

Next steps include birthplace geocoding, historical timezone and daylight-saving conversion, optional true-solar-time correction, independent chart benchmarks, consent-first accounts, encrypted storage, deletion controls, and matching signals based on friendship intent, values, distance, and boundaries.

Built With

• react
• typescript
• vite