Sentiment Dashboard
Upload a CSV of customer reviews or support messages and get a sentiment breakdown: positive, negative, neutral percentages, a trend chart over time, and a flagged list of the most negative messages. Uses VADER for fast rule-based sentiment, with optional LLM pass for ambiguous messages.
Throughput
~2,000 msg/sec
Accuracy
91% (hybrid)
Max batch
50,000 messages
Data stored
Never
How it works
Analysis Pipeline
VADER runs first (free, instant, no API). Messages with compound score between -0.15 and +0.15 (ambiguous zone) are optionally passed to an LLM for nuanced classification. Keeps costs low — typically only 10-15% of messages need the LLM.
Accuracy & Tradeoffs
VADER only: 84% accuracy at ~2k msg/s. VADER + LLM hybrid: 91% accuracy at ~800 msg/s effective. LLM only: 93% but at 5x the cost. The hybrid approach gives 91% accuracy at 1/5th the cost of all-LLM — the 2% difference is negligible for broad sentiment categories.
Security
- Zero data persistence — CSVs processed in-memory and discarded
- PII redaction before any LLM calls
- Input limits: 20 MB, 50k messages, 5k chars/message
- Rate limiting: 3 analyses/min per IP
- CSV injection prevention: formula characters stripped
Want something like this built for your business?
I'll look at your problem, figure out the right approach, and ship working software. No slideshows.
Book a free consultation