Andromeda
Professional-grade local AI news monitoring system
Monitoring system that captures, transcribes, and analyzes live financial news streams using local Whisper models and LLMs. Turns raw audio into structured intelligence you can actually search. Runs entirely on local hardware, so nothing leaves your machine.
Financial news moves fast, but analysts lose time when live audio, transcripts, summaries, and searchable context live in different tools or depend on cloud services that raise cost and privacy concerns.
Defined the product approach and built the core transcription, enrichment, and retrieval workflows for a local-first monitoring system.
System Constraints
Continuous audio ingestion with near-real-time transcription
Local execution to avoid sending sensitive monitoring data to third parties
Searchable retention across transcripts, entities, and themes
Hardware-aware performance tradeoffs on non-datacenter infrastructure
What I Built
A local AI monitoring pipeline using Whisper-class transcription, diarization, semantic search, and lightweight LLM enrichment.
A processing split between always-on extraction and heavier downstream analysis so the system could stay responsive.
A searchable intelligence layer that made live and historical news equally usable inside the broader Parallax workflow.
Outcome / Takeaway
Converted raw broadcast audio into searchable, structured intelligence that could feed analysts and downstream products.
Preserved privacy and cost control by keeping the stack local-first.
Extended Parallax from market and OSINT signals into spoken media and live commentary.
The product win came from making local AI feel dependable and operational, not experimental. Reliability mattered as much as model quality.