What the work actually looks like.

The problem

Over $1M/year spent on manual reconciliation across policy admin, billing, and claims systems. 13 full-time employees matching transactions by hand. Error rates climbing, process couldn't scale.

What we did

Audited the full workflow. Deployed a locally-hosted LLM fine-tuned on 2 years of historical reconciliation data. Built an end-to-end pipeline: ingest, classify, match, resolve, report. Confidence scoring routes edge cases to human review.

The problem

$18K/month on cloud AI APIs for document classification. Every customer document processed through third-party servers. Compliance flagged the privacy risk. Costs growing 15% quarter over quarter.

What we did

Assessed the pipeline, fine-tuned a 13B model for financial document classification. Deployed on a single GPU server ($22K) in their server room. Built a drop-in REST API matching their existing integration — zero code changes on their side.

The problem

3 weeks every quarter preparing regulatory reports. Data from 6 systems, cross-referenced against requirements, narrative summaries, formatting. Consumed the entire compliance team.

What we did

Automated ETL from all 6 source systems. LLM trained on 3 years of approved report language generates narrative sections. PowerBI dashboard for the compliance team to validate. Full audit trail.