Production systems & client work

The problem

A mid-sized professional services firm processed 200+ documents per week manually — extracting key data, classifying document types, and routing to the right teams. The process consumed ~40 hours of staff time weekly and was error-prone under volume spikes.

The approach

I designed and built an AI-powered document processing pipeline using a combination of OCR, language model extraction, and rule-based routing. The system classifies incoming documents, extracts structured data into their existing tools, and routes items to the appropriate team with confidence scoring.

What changed

• •Document processing time reduced by 75%
• •Manual review limited to edge cases and low-confidence extractions
• •Staff reassigned ~30 hours/week to higher-value work
• •Error rate dropped from ~8% to under 2%

Phil's role

Architecture design, model selection, pipeline development, integration, and team training.

The problem

A growing consulting firm had 10+ years of project documentation, proposals, deliverables, and internal research spread across shared drives and legacy systems. Finding relevant past work took senior consultants 30–60 minutes per search, and junior staff often couldn't find it at all.

The approach

I built a secure internal RAG-based assistant that indexes their document library and allows natural language search across all historical project materials. The system runs on their infrastructure with no external data exposure, using an open-source language model for privacy compliance.

What changed

• •Average information retrieval time dropped from 30–60 minutes to under 2 minutes
• •Junior staff onboarding time reduced significantly
• •Proposal preparation accelerated by reusing relevant past work
• •Institutional knowledge became accessible instead of locked in senior heads

Phil's role

Requirements gathering, RAG architecture design, open-source model deployment, document pipeline development, and user training.

The problem

A financial services operations team produced weekly and monthly reports by manually pulling data from multiple systems, formatting spreadsheets, writing narrative summaries, and reviewing for accuracy. Each reporting cycle consumed 15–20 hours across the team.

The approach

I built an automated reporting pipeline that pulls data from their existing systems, generates structured reports with AI-written narrative summaries, and flags anomalies for human review. The system drafts complete reports that analysts review and approve rather than create from scratch.

What changed

• •Report generation time reduced from 15–20 hours to 2–3 hours of review
• •Reporting consistency improved across cycles
• •Analysts focused on interpretation and decision-making instead of data assembly
• •Anomaly detection caught issues that manual review had missed

Phil's role

Data pipeline architecture, AI summarization system design, integration development, and workflow redesign.

The problem

A mid-market company with ~200 employees wanted to adopt AI across their operations but had no internal AI expertise. They'd received pitches from multiple vendors and agencies but couldn't evaluate the options or determine where to start.

The approach

I conducted a structured AI opportunity audit — mapping their core workflows, identifying automation candidates, evaluating build vs. buy options, and producing a phased implementation roadmap with realistic cost and timeline estimates.

What changed

• •Leadership had a clear, prioritized roadmap instead of vendor confusion
• •Three high-value projects identified and scoped for Phase 1
• •Two low-value vendor proposals correctly declined, saving $150K+ in misdirected spend
• •Internal team aligned on AI strategy and realistic expectations

Phil's role

Operational assessment, opportunity mapping, architecture advisory, vendor evaluation, and roadmap delivery.