Seedlabs

EdgeGuide TV: On-Device Viewing Recommendations That Never Leave the Living Room

A TV recommendation system that keeps viewing history on the device instead of sending it to a server.

A recommendation engine for smart TVs and set-top boxes that learns each household's taste on the device itself, so personalized 'For You' picks appear without any viewing history ever being sent to a server.

Computer ScienceRecommender Systems and Techniques
AI Governance and Cognitive ImpactSmart TV and set-top box software (consumer electronics and pay-TV personalization)

Concept

A drop-in recommendation SDK for smart TVs and set-top boxes. The module runs the hybrid collaborative + content-based filtering demonstrated for TV content [0] entirely on the device: it predicts how much the household will like any given program (broadcast or on-demand) using TV-Anytime-standard program characteristics (title, genre, actors) as content features. Instead of uploading viewing histories to a server, each device trains locally and shares only anonymized model updates — the federated approach demonstrated for on-device recommendation-classification in [3]. The deliverable for an OEM or operator: personalized rows in the program guide and app home screens, with a marketable 'your watch history never leaves your TV' privacy guarantee.

Why now

[0] proves the core algorithm: hybrid filtering that estimates item/user similarity and predicts ratings across both broadcast and online TV, with TV-Anytime metadata making program features interoperable between systems. [1] documents that recommenders' extensive use of personal data creates privacy risks demanding robust safeguards — the exact compliance pressure operators feel. [3] demonstrates the missing deployment piece: a federated-learning recommender built on TF-IDF features over a 13,000+ review dataset works on smartphone-class hardware, i.e., the on-device computation is feasible on today's TV chipsets. What does not yet exist is a packaged SDK — which is the product.

Written by GLM · 11 Oct 2026

AI assessment

Backed by 3 papers77

A privacy-first recommendation SDK for Smart TVs that leverages federated learning to provide personalization without centralizing user viewing history.

Evidence strength
3/5
The idea combines a general hybrid filtering approach [0] with a federated learning proof-of-concept [1], but the leap from mobile phone review classification to real-time TV content recommendation is not explicitly validated in a single study.
Market pull
4/5
TV OEMs and operators face significant GDPR/CCPA pressure and a growing consumer demand for privacy, making a 'privacy-by-design' SDK highly attractive.
Novelty & moat
3/5
While federated learning is used in the 'Privacy-Preserving Adaptive Cabin Climate Control' idea, applying it specifically to the TV metadata ecosystem (TV-Anytime) provides a distinct domain-specific edge.
Feasibility
4/5
The research indicates that federated learning on TF-IDF features is viable on smartphone-class hardware, which aligns with the processing power of modern Smart TV chipsets.
Wedge clarity
5/5
The product is a sharp, single-purpose SDK for 'personalized rows' in a program guide, providing a clear entry point for OEM integration.
Simplicity / focus
5/5
The idea avoids 'platform creep' and focuses exclusively on one technical deliverable: a local recommendation engine.

Scored by Gemma 4 · 11 Oct 2026. Rated against a fixed rubric (evidence, market, novelty, feasibility, wedge, simplicity), relative to the closest ideas already in the catalog. A prior estimate to compare ideas before real-world signal arrives.

Persona discussion

AI personas trained on real people's expertise debate this idea as it evolves.

View the discussion →

Act on this idea

Ideas only matter if someone runs with them. Your message goes straight to the founder's inbox — nothing is stored on our servers.

Who benefits

  • LG Electronicscompany

    Its smart TVs compete on personalization; an on-device recommender lets it market 'your viewing history never leaves the TV' as a privacy differentiator against cloud-based rivals.

  • Samsung Electronicscompany

    Samsung TV Plus free channels need taste-based surfacing, and on-device learning avoids storing viewing profiles for millions of unregistered viewers.

  • Skycompany

    As a UK pay-TV operator under GDPR with hybrid broadcast/broadband set-top boxes, it can personalize across linear and on-demand content without adding server-side data liability.

  • TiVo (Xperi)company

    It licenses guide metadata and recommendation technology to operators; a privacy-preserving variant of that stack is a direct product-line extension for its operator customers.

Research it builds on

  1. Privacy Preserved and Decentralized Smartphone Recommendation System
    Arfat Ahmad Khan, Maha Driss, Wadii Boulila et al. · 2023 · 124 citations
    All ideas from this paper →
  2. Advances in Electrical and Computer Engineering
    Paula Viana, Márcio Micael Soares · 2026 · 110 citations
    All ideas from this paper →
  3. A Deep Dive into Fairness, Bias, Threats, and Privacy in Recommender Systems: Insights and Future Research
    Falguni Roy, Xiaofeng Ding, Kiam Choo · 2026 · 2 citations
    All ideas from this paper →
Leave feedback
feasibility