Orlik

The Rohde & Schwarz challenge at the European Defense Hackathon in Munich focused on extracting intelligence from intercepted audio communications.

Operators face significant challenges: noisy recordings, abbreviated language, codewords, and the need to extract meta-information beyond just transcription. The goal was to analyze not only what people say, but how they say it, detect stress levels, estimate speaker hierarchy, and identify environmental context through background sounds.

We designed Orlik as a comprehensive audio intelligence solution that processes intercepted communications through multiple automated stages, generating actionable intelligence summaries for operators.

• Audio processing pipeline: Automated transcription with speaker diarization to identify who spoke when, even in noisy conditions with multiple speakers and overlapping speech.
• Intelligence extraction: AI-powered analysis to extract locations, relationships between speakers, threat severity assessment, military ranks and units, and key entities mentioned in conversations.
• Background audio classification: Detection and categorization of environmental sounds (vehicle types, location indicators, ambient noise) to provide situational awareness.
• Interactive intelligence dashboard: Real-time visualization with waveform analysis, speaker analytics, and interceptions overlayed on a map for enhanced visual clarity.

The application features a React-based desktop interface with a Python backend handling the computationally intensive audio processing.

The audio pipeline performs voice activity detection, speech recognition with automatic translation, speaker identification, and background noise classification. Results flow into the text analysis stage which uses language models to extract structured intelligence metadata.

All results persist to a real-time database with JSONB fields enabling complex queries on extracted intelligence.

Completed from scratch in under 48 hours, Orlik was selected to present before defense industry professionals from an initial pool of 56 hackathon projects.

The application successfully demonstrated end-to-end intelligence extraction from raw audio to actionable insights, with a polished interface showcasing real-time processing, automated analysis, and comprehensive visualization of results.