What it is
Redreem (live at redreem.ai) is a travel intelligence product that compares cash fares against award availability and ranks options by what the user can actually act on — affordability, availability, and points balance — rather than headline price.
The system uses an LLM cascade for itinerary parsing and recommendation, with API orchestration over Duffel (cash fares), Seats.aero (award availability), and Resend (transactional email).
Why it matters as a portfolio piece
Redreem is the consumer-facing complement to Transferithm. Where Transferithm demonstrates calibration on a low-volume, public-graded prediction problem, Redreem demonstrates feedback-loop design on a high-volume consumer surface — every booking outcome becomes a labelled training signal.
It’s also the project where I get to make and defend product judgement calls: ranking hierarchies, dark-pattern boundaries, monetisation paths that don’t compromise editorial independence.
Status
Layer 2.5 ~95% complete. Reverse search MVP in progress. Affordability-first ranking is a hard rule, not a score input — see the writing log for the rationale.