Radar
Protect email,
comment, and registration
flows with a spam protection API.
RADAR uses a smart weighted spam scoring engine. We extract normal words from all text processed, filter and fact check it with the localized language and use a scoring engine to combine heuristics into final score
FAQs
Traditional filters rely heavily on static keyword blocking, which spammers easily bypass. Our engine uses a dynamic weighted scoring system. It extracts text, analyzes standard language patterns, filters anomalies, and fact-checks content against localized linguistic data. Instead of a simple “yes or no” flag, it combines multiple heuristics to give you a highly accurate, nuanced probability score.
The engine evaluates incoming text through multiple independent lenses (heuristics), such as syntax anomalies, sender reputation, and localized language patterns.
Each test is assigned a specific “weight” based on its reliability.
The engine aggregates these weighted inputs to calculate a final consolidated score.
You can set your own thresholds to block, flag for review, or allow content based on this score.
Spam looks different in every language and region. Slang, common idioms, and phrasing vary wildly. Our engine doesn’t just translate text; it analyzes it against the specific rules and nuances of the local language context. This drastically reduces false positives caused by regional dialects or informal business language.
The engine cross-references extracted text and claims against verified localized datasets to detect known phishing templates, fraudulent business pitches, and common scam setups. If a piece of text closely mimics a known malicious campaign or contains verifiable misinformation (like a fake lottery claim), its spam score increases accordingly
Yes. Every business has a different risk tolerance. Our engine allows you to customize the heuristic weights via our dashboard or API. For example, if your platform is highly sensitive to financial scams, you can increase the weight assigned to currency and wire-transfer heuristics.
Not at all. The text extraction and processing happen in real-time within a secure, encrypted pipeline. We isolate the standard linguistic patterns and words required for scoring, and we do not store the PII (Personally Identifiable Information) or the contents of the processed text post-analysis.
Because our engine extracts normal words and filters them against localized language models, it excels at spotting “leetspeak” (e.g., replacing ‘E’ with ‘3’) or intentional misspellings meant to trick basic filters. The system recognizes the underlying intent and penalizes the text for obfuscation tactics.
Despite running multi-layered heuristic checks and language filtering, the engine is optimized for high-throughput environments. Average response times are sub-100 milliseconds, making it seamless for real-time applications like chat platforms, form submissions, and email routing.
API Response
Simple API
Our API delivers clear, actionable intelligence in real time. For every text chunk processed, the engine returns a consolidated score between 0.00 and 1.00, an automated classification tag (spam, abuse, blocked, or malicious), and a detailed breakdown of the underlying signals. This transparent telemetry shows you exactly why a piece of content was flagged—tracking everything from token counts and uppercase ratios to localized spam keyword density—allowing your team to seamlessly automate moderation workflows.
{ "score": 0.95, "classification": "spam", "signals": { "token_count": 10, "unknown_word_ratio": 0.1, "spam_keyword_score": 0.61, "uppercase_ratio": 0.34, "exclamation_density": 0.3, "url_count": 1 } }