ShieldLine
Mockup hub · clickable prototype
An NLP-powered classification API scores every inbound SMS as ham or spam in under 100ms using a Decision Tree model trained on TF‑IDF features — quarantining malicious messages before they reach a phone, while a feedback loop and monthly retraining pipeline keep it adapting to new spam vocabulary.
96.59%
Test accuracy (Decision Tree)
93.50%
Test precision on spam class
<100ms
P99 classification latency
13.5%
Spam rate across SMS traffic
Four personas from the PRD's user stories, each with their own corner of the product.
CISO / Security Manager
Wants visibility into the SMS threat landscape and confidence the filter is working. — Story 1, 4
Dashboard · Alerts · Live Feed
Business Operations Manager
Needs a blocked legitimate message released fast, and a way to flag mistakes. — Story 2
Report a Misclassification · Quarantine
Telecom / Platform Engineer
Integrates the classification API into routing infrastructure and watches its health. — Story 3
Dashboard · Settings (API keys)
ML / Review Analyst
Labels borderline messages within SLA and oversees model retraining and rollouts. — Flow 5
Review Queue · Model & Retraining
The five user flows from PRD §3.2 — each maps to one or more screens below.
Flow 1 — Real-time classification
SMS in → preprocess → TF‑IDF → Decision Tree → label + confidence → quarantine or deliver → audit log. <100ms
Watch the live feedFlow 2 — Misclassification feedback loop
User flags a wrong call → feedback record created → analyst confirms → correction queued for retraining.
Open the report flowFlow 3 — Monthly model retraining
Pull feedback + new messages → retrain → evaluate vs. holdout → promote if precision ≥93% & recall ≥90%, else hold.
See the retraining pipelineFlow 4 — Admin monitoring & alerting
Dashboard tracks spam rate, false-positive rate, accuracy, latency → alerts on spikes or degradation → weekly report.
View the dashboardFlow 5 — Human-in-the-loop review (low confidence)
Messages scored 40–60% confidence are routed to an analyst queue (not auto-blocked) → reviewed within 4 business hours → label applied and added to the training queue immediately.
Open the review queueOrdered like a storyboard — open index.html any time to come back here.
Security-console authentication for staff users.
North-star metrics, spam-rate trend, live alert banner. FR7, Story 4
Streaming view of every message scored in real time. Flow 1, FR1, FR2
Held spam, 30-day retention, one-click release. FR3, FR4, Story 2
40–60% confidence messages awaiting an analyst label. Flow 5, FR6
Hash, audit trail, and the top contributing TF‑IDF terms. §4.4 Explainability
End-user portal for flagging a wrongly blocked message. Flow 2, Story 2
Spike detection, weekly security report, CSV export. FR7, Story 4
4-model comparison, accuracy gate, version history. Flow 3, FR5, §5.1
Thresholds, sender allowlist, API keys — gated by approval. §4.3