John de Graft-Johnson

Experience —

Eight years across data engineering, applied ML, and AI platform delivery — recently focused on evaluation frameworks, fine-tuning, and responsible-AI tooling.

1. 01

   AI Engineering Lead · Swain Solutions LLC

   Oct 2025 – Present

   Washington, D.C.

   • › Shipped production agentic LLM systems end-to-end for regulated customers — a 32-tool MCP server for AI proposal intelligence, a Clinical Decision RAG grounded in NICE / SNOMED / FHIR R4, and a LangSmith-instrumented chat inspector with 7 evaluators — embedding with stakeholders to translate ambiguous requirements into shipped, customer-facing surfaces.
   • › Owned a 9-service Azure-native LLM platform plus a 37-Lambda AWS counterpart: TypeScript / Next.js front-ends on Vercel, FastAPI handlers, Cosmos DB / ADLS medallion data (dbt-duckdb bronze → silver → gold), sub-1ms CUDA GPU inference (BART/BERT); federated learning across 217 tickers with a live options calibration model over 19K+ rows.
   • › Fine-tuned open-weights models (Gemma family) with LoRA and preference-optimization workflows; built deterministic judges, paired LLM auditors, and weighted composite scorecards covering fairness, factuality, and compliance — instrumented with inspect_petri red-teaming, Fairlearn audits, and LangSmith observability so customers see evidence, not assertions.
   • › Embedded a 13-point Responsible-AI framework (HIPAA per NIST SP 800-66r2, CMMC, MHRA GMLP) into CI gates, IaC, and Solana-anchored audit trails; every customer deployment ships with a reproducible evidence pack rather than a checklist. Mentored engineers on cross-domain ownership (data + ML + infra + frontend) and customer-facing delivery standards.
   PythonTypeScriptNext.jsLangChain / LangGraph / LangSmithMCPRAGLoRA / SFTvLLMAzureAWS
2. 02

   Product Analytics Manager · W.R. Grace

   Jan 2024 – Oct 2025

   Columbia, MD

   • › Spearheaded enterprise commercial and volume forecasting systems powering $700MM+ revenue planning, capital allocation, and scenario-based performance modeling; embedded with Finance, Operations, and executive stakeholders to translate ambiguous business questions into decision-ready forecasts.
   • › Shipped a forecast-accuracy measurement and reconciliation framework — instrumented as a production evaluation harness across business units — reducing forecast-to-actual variance by 18%.
   • › Set analytics platform strategy and cloud data architecture (AWS / Azure) for enterprise-scale modeling and demand planning; advanced ML and predictive-modeling frameworks for product performance, adoption velocity, and market risk; owned data integrity and enterprise analytics standards.
   ForecastingScenario ModelingML / PredictiveAWS / AzureGovernance
3. 03

   Senior Data Analyst · W.R. Grace

   Aug 2022 – Jan 2024

   Columbia, MD

   • › Directed enterprise revenue-intelligence program across SAP and Salesforce ecosystems using SQL-driven analytics, delivering visibility into $600MM+ monthly performance and growth signals to executive stakeholders.
   • › Deployed statistical forecasting and performance models to production, reducing planning cycle time by 25% and stabilizing revenue predictability; engineered centralized pipelines integrating SAP, Salesforce, and operational datasets.
   • › Shipped executive-facing KPI visualization frameworks (Tableau, Power BI); applied segmentation and cohort modeling to surface evidence-based signals for go-to-market prioritization.
   SAP / SalesforceSQLStatistical ForecastingTableau / Power BISegmentation
4. 04

   Data Engineer · W.R. Grace

   Jul 2019 – Aug 2022

   Baltimore, MD

   • › Led data integrity, governance, and compliance programs across regulated reporting environments; established standards and audit frameworks later scaled at the enterprise level.
   • › Designed Power BI dashboards with automated anomaly detection and process-deviation monitoring across multi-site operations; evidence-based interventions reduced system downtime by 28% and sustained 99% product quality.
   • › Formulated capacity-planning algorithms and volume-forecasting models in SQL and Python that reduced bottlenecks by 20% and stabilized production planning; refined predictive-modeling frameworks through algorithm tuning and feature engineering.
   • › Raised data accuracy by 14% through systematic predictive modeling, outlier remediation, and root-cause analysis; maintained scalable data models supporting performance monitoring and early volume forecasting.
   Power BIAnomaly DetectionPredictive ModelingGovernanceSQL / Python