Beginnerstable00_/_BEGINNER

Hate Speech Detector

A single-agent CrewAI pipeline that analyzes text for hate speech or offensive language, classifying it with confidence and reasoning.

Py_3.12CrewAIOpenRouterVIEW SOURCE

# INPUT_PROMPT // HATE_SPEECH

SCHEMA: V.1.0

>>>TEXT_INPUT

People from that race are all untrustworthy and should not be allowed to work in public jobs.

system@agent-archive:~$

IDLE

OUTPUT_LOG:

[IDLE] Awaiting input. Press RUN DEMO to initialize the pipeline.

LIVE_EXECUTION_LOG // hate-speech-detector0 EVENTS

[STREAM] No active execution. Trigger a run to populate the log.

/SYSTEM_INTERNALS

ARCHITECTURE

SYSTEM_ARCHITECTURE // CREWAI_SINGLE_AGENT_PIPELINE

INPUT

field: text (textarea)

Raw text string to analyze for hate speech or offensive language

CREW

agents: [hate_speech_detector]

tasks: [hate_speech_detection_task]

verbose: false

crew.kickoff(inputs={"text": Text})

AGENT

role: Hate Speech Detector

goal: Analyze text and identify hate speech

backstory: "You are a Hate Speech Detector for Twitter who understands details very well..."

LLM

model: gpt-4o

provider: OpenRouter

TASK

description: 5-step analysis pipeline

1. Read text carefully

2. Identify targeting language

3. Look for threats / dehumanizing language

4. Evaluate context

5. Make objective classification

expected_output: structured analysis

STRUCTURED OUTPUT

classification: hate speech | no hate speech

confidence: high | medium | low

targeted_group: race, gender, ...

key_phrases: [extracted phrases]

reasoning: 2-3 sentence explanation

/BACKEND_CONTRACT

PLUGGABLE RUNTIME

This demo serves a hardcoded fixture today. Set NEXT_PUBLIC_USE_REAL_BACKEND=true and the same UI will POST /api/beginner/hate-speech-detector to a real CrewAI service - no UI change required.

/TYPED_INTERFACE

runProject(HATE_SPEECH, {
  text: string,
}) -> markdown