SCOREData & productivity
Soar Real-Time Social Data Index
John Tadeusiak@privateoracle
Soar indexes and aggregates social data to provide users with real-time context on companies they trade, utilizing a Jev-based pipeline for improved efficiency and performance.
first @typesafeai implementation live in production on @TrySoar.
200k requests.
$4 total cost.
~200ms latency.
soar indexes and aggregates social data (tweets, rss) so users have real-time context on the companies they trade.
previously we ran a stack of hf models on every piece of incoming data:
sentiment analysis (3 separate models aggregated)
ad detection
categorization (funding rounds, growth, etc)
subject extraction
it was expensive, slow, and choked under high load.
re-indexed the entire pipeline with jev:
sentiment: 3 models collapsed into 1 with better scoring
ad detection + categorization: merged into 1 zero shot
subject extraction: massive reduction in false positives
run stats:
cost: $4
latency: ~200ms per request
requests: 200k
i posted yesterday that jev was the first model i've been excited about in a while. cheap, fast, and structured out of the box, which makes dropping it into production code effortless.
this was just the easiest baseline implementation we could run. building out much deeper architectures with it next.