๐
Brand-specific encoding tables
We build a decoder for each manufacturer's serial format from official spec sheets and verified production data โ not reverse-engineered guesses.
๐ฌ
AST-sandboxed Python decoders
Each decoder runs in an isolated sandbox. New brand decoders are generated by AI, then validated against known serials before going live.
๐
Confidence scoring
Every decode returns a confidence score โ HIGH, MEDIUM, or LOW. If we can't decode a serial, we return null rather than fabricate a date.
๐
Automatic learning
When our AI cracks a new brand's serial format through pattern analysis, it's tested, validated, and added to the decoder library automatically.
โก
Millisecond decode
Serial decoding is deterministic โ not an LLM call. Once the decoder exists, it runs in <1ms. Latency comes from the rest of the lookup pipeline.
๐
International format support
Handles serials from US, Canadian, European, and Asian manufacturing plants โ including multi-factory brands like Samsung and LG.