Launch HN: Lumona (YC W24) – Product search based on Reddit and YouTube reviews

Hacker News Discussion ↗

Overall reception & use cases

  • Many commenters like the idea, noting they already do “Google + Reddit” manually and want a cleaner interface for this.
  • Others say current limitation to skincare makes it not yet useful for them; they’d want it for supplements, electronics, vacuums, yoga mats, restaurants, TV shows, games, and general “what should I do” queries.
  • Some see it as a promising example of a focused, domain-specific LLM product, as opposed to generic chatbots.

Trust, shilling, and data quality

  • Strong concern that Reddit and YouTube are already heavily astroturfed: paid posts, high‑karma account markets, upvote networks, and undisclosed sponsorships.
  • Several argue a product that formalizes “Reddit reviews” will further incentivize spam and push genuine conversation into closed communities.
  • Others note Reddit is still often trusted more than random blogs, but acknowledge this is a low bar.
  • Suggestions include: shill graphs to downweight suspicious accounts, analyzing user histories and subreddit bias, detecting GPT‑like text and AI videos, and separating brand‑fan subs from general ones.
  • There is skepticism that basic filters (AutoModerator removal, relying on YouTube sponsorship flags) are sufficient or realistic.

Monetization and incentives

  • Multiple people warn that an affiliate model tends to corrupt rankings over time, pushing high‑commission or sponsored products rather than “best” ones.
  • Comparisons are made to Wirecutter and Consumer Reports; some propose a subscription or ad‑supported model as more trustworthy.
  • Past builders of similar tools report that aggregators are hard to monetize and affiliate programs are shrinking.

Scope, UX, and performance

  • Confusion arises because branding suggests general product search, but only skincare works; non‑skincare queries can hang or misbehave.
  • Users report slow responses, janky scrolling, inconsistent “top product” results, unclear navigation (vertical vs horizontal swipes), and ambiguous arrows vs distributor lists.
  • Desired features: side‑by‑side product comparison, clearer card counts, ingredient databases with cited sources, and multimodal input (e.g., upload skin photos).

Medical and skincare concerns

  • Some criticize reliance on anecdotal Reddit skincare advice and omission of key prescription treatments.
  • There is debate about side effects of tretinoin vs benzoyl peroxide and the importance of lifestyle factors (diet, sleep, etc.), with calls for surfacing non‑product solutions too.

Technical and data sourcing

  • The team uses Reddit dumps (monthly) instead of the official API and fine‑tunes a small open model on GPT‑4 outputs for cost control.
  • There is a brief discussion about whether such fine‑tuning conflicts with OpenAI’s terms; interpretation is acknowledged as uncertain.