How asimila Works
From raw political discourse to structured, citable intelligence — a four-stage pipeline built for accuracy, provenance, and scale.
asimila operates a continuous intelligence pipeline that transforms raw political discourse from Chilean legislative institutions and media channels into structured, evidence-backed intelligence. Each stage is designed for accuracy, auditability, and scale — from automated source capture through AI transcription and entity extraction to secure team delivery. Every piece of output intelligence maintains a citation provenance chain back to the original source.
Curate Sources
asimila continuously monitors 50+ curated source channels spanning Chilean legislative institutions, media outlets, and public discourse platforms. Source types include HLS live streams from Senate plenary sessions, institutional session pages from Congress committees, RSS feeds from legislative publications, REST APIs from government data portals, and direct media file endpoints. Polling daemons check each channel on configurable intervals, detecting new sessions as they are published. When a new session is identified, the platform automatically downloads and normalizes the source media — whether audio, video, or document — and queues it for transcription. Channel configurations define provider-specific logic for URL resolution, metadata extraction, and deduplication to prevent reprocessing of previously captured sessions.
- 50+ channels spanning Senate, Congress, and media sources
- HLS streams, RSS feeds, APIs, and direct media capture
- Automated polling with deduplication and session detection
Transcribe and Attribute
Captured audio is transcribed using AI speech-to-text with speaker diarization, producing utterance-level segments each tagged with speaker turn boundaries and timestamps. The transcript then passes through a transformation stage that applies domain-specific corrections — normalizing legal references, bill numbers, institutional names, and political terminology for Chilean legislative context. Speaker labels from diarization are resolved against a canonical speaker registry through phonetic matching and entity resolution, linking anonymous speaker turns to verified identities. The transformation pipeline also handles number normalization, acronym expansion, and vocabulary corrections using a curated key-terms registry. The result is a speaker-attributed transcript where every utterance is linked to a verified entity with consistent naming across sessions.
- Speaker diarization with entity resolution against canonical registry
- Domain-specific corrections for Chilean legislative terminology
- Phonetic matching and key-terms normalization for accuracy
Extract and Cross-Reference
The extraction pipeline applies LLM-based analysis to speaker-attributed transcripts, identifying entity mentions, key topics, proposals, voting records, action items, legal references, and procedural motions from each session. Every extracted item carries citation provenance linking it to the specific utterance, speaker, and timestamp from which it was derived. Extraction operates on windowed transcript segments to maintain context while respecting token limits, with checkpoint recovery to handle long sessions reliably. Following extraction, the consolidation stage merges results across segments — deduplicating entities, reconciling topics, and building cross-references. Bills mentioned in multiple sessions are linked into a unified tracking timeline. Speaker positions are aggregated across appearances to build longitudinal records.
- Entity, topic, and bill extraction with citation provenance
- Cross-session consolidation with deduplication and linking
- Windowed processing with checkpoint recovery for reliability
Deliver to Your Team
Structured intelligence is delivered through secure, role-based team workspaces where analysts can search, filter, and monitor political developments. Workspaces provide session-level analysis views with speaker-attributed transcripts, extracted entities, topic summaries, and cross-source references. Teams can configure monitoring alerts for specific bills, entities, or topics to receive notifications when new discourse matches their criteria. Intelligence can be exported in multiple formats — PDF and DOCX for formatted reports and briefs, Markdown for editorial workflows, and JSON for integration with external systems and databases. Every export preserves the full citation provenance chain, ensuring that downstream consumers can trace any claim back to its source. Organization-level access controls and AES-256-GCM encryption protect sensitive analysis and private source material.
- Role-based workspaces with search, filtering, and monitoring
- Export as PDF, DOCX, Markdown, or JSON with full citations
- AES-256-GCM encryption and organization-level access controls