Model Sommelier — Smallest Sufficient AI
Discovery Lens
C Combination Innovation
Two separate worlds finally connect — and the intersection is a product
One-Liner
An API middleware that automatically routes each AI request to the smallest (cheapest, fastest) model that can handle it at the required quality level.
Kill Reason
RouteLLM (open-source, Lmsys) and LiteLLM already provide model-routing middleware, and multiple well-funded startups are building commercial versions. Without proprietary usage data at massive scale to train better routing heuristics, this remains a commodity middleware layer that enterprises can self-host for free.
What do you think?
Related ideas you can explore free:
killed: The FDA approval barrier for health screening claims is prohibitive for a startup — without regulatory clearance, the product must avoid medical language, which hollows out the core value proposition. Apple and Google are building voice and sensor health monitoring directly into their platforms, and they have the regulatory resources to navigate FDA that a startup cannot match.
killed: Consumer-grade distributed inference faces an unsolvable latency problem: interactive AI workloads require sub-300ms round trips, but coordinating across residential internet connections with variable uptime makes this physically implausible. The market need is already served by cloud inference providers (Together.ai, Fireworks.ai, Groq) with low latency and no coordination overhead, and hardware accelerators (Groq LPU, Cerebras) are collapsing inference costs faster than consumer compute aggregation could.
killed: The Sense Home Energy Monitor and Google Nest already provide appliance-level consumption insights to millions of homes via hardware and platform integration. Utility-side analytics platforms (Oracle Utilities, Itron) serve the B2B market. The Green Button API dependency limits reach to approximately 60% of U.S. utilities, and the incumbents already hold the integrations, brand trust, and hardware relationships needed to serve both consumer and enterprise segments.