Production AI/ML engineering — geospatial intelligence, LLM orchestration, RAG pipelines, and voice biometrics. Built and deployed on Google Cloud Platform. Open-source repos and full case studies at danielflugger.com.
Automated Geospatial Property Verification
Production ML pipeline for automated geospatial property verification. Cross-references property records, parcel boundaries, and document-claimed attributes against live geospatial data. Combines spatial analysis with ML-based anomaly detection to flag discrepancies between recorded and actual property characteristics — serving use cases in real estate due diligence, insurance underwriting, and municipal assessment.
Vertex AI model training and serving, geospatial feature engineering, anomaly detection on parcel and attribute data
PostGIS spatial queries, BigQuery analytical pipelines, automated data ingestion from public records and GIS sources
Production RAG for Engineering Document Workflows
Production RAG system for engineering document workflows. Ingests multi-source documents — RFPs, proposals, inspection reports — and generates structured outputs using grounded LLM retrieval via Vertex AI. Designed for domains where document accuracy has financial and legal consequences: construction, procurement, and compliance review.
Retrieval-augmented generation, document embeddings, semantic chunking, grounded generation with citation tracking
FastAPI serving layer, Pydantic schema validation, vector search with Vertex AI, Cloud Run deployment
AI-Driven Procurement Intelligence Platform
LLM-powered procurement intelligence across several hundred vendor sources. Aggregates supplier data, applies Gemini-powered semantic scoring to match vendors against project briefs, and generates structured comparison reports. Built to replace manual vendor research in construction, manufacturing, and enterprise procurement workflows.
Gemini API semantic scoring, LLM-powered entity extraction, embedding-based vendor matching, structured report generation
Multi-source data aggregation, BigQuery analytical warehouse, automated ingestion pipelines
Voice Biometric Identity Platform
Zero-knowledge privacy architecture for voice-based identity verification. Five-layer verification stack with client-side encryption, blockchain timestamping, and a full ML pipeline for voiceprint analysis. Processes acoustic features through Resemblyzer, Wav2Vec2, Parselmouth, and librosa for multi-dimensional speaker verification — without ever storing raw audio server-side.
Speaker embedding models (Resemblyzer, Wav2Vec2), acoustic feature extraction (Parselmouth, librosa), multi-layer biometric scoring
Zero-knowledge design, client-side encryption via Web Crypto API, blockchain timestamping via OriginStamp
Automated Regulatory Document Analysis for Financial Services
Multi-agent LLM system for a financial services firm that automates regulatory compliance review across SEC filings, internal policy documents, and audit trails. Orchestrates specialized agents — extraction, cross-reference, gap analysis, and report generation — through a Vertex AI pipeline. Reduces manual compliance review cycles from weeks to hours while maintaining full audit traceability.
Engagement started October 2025. Client details under NDA.
Multi-agent orchestration, Gemini-powered document understanding, entity extraction, semantic similarity for policy cross-referencing
Vertex AI Pipelines, Cloud Run microservices, Firestore for agent state management, BigQuery for audit logging
Time-Series Anomaly Detection for Industrial IoT
Predictive maintenance system for a mid-market logistics operator ingesting telemetry from fleet and facility IoT sensors. Combines time-series anomaly detection with a classification model that predicts equipment failure windows, routing maintenance alerts through a priority scoring system. Designed to reduce unplanned downtime and shift maintenance from reactive to condition-based scheduling.
Engagement started December 2025. Client details under NDA.
Time-series anomaly detection, failure prediction classification, sensor feature engineering, model retraining pipelines
IoT telemetry ingestion via Pub/Sub, BigQuery streaming inserts, Vertex AI batch and online prediction serving
If you're working on an AI/ML system, data platform, or GCP deployment and need an engineer who ships — let's talk.
Start a Conversation →