Jerome Hardaway

Summary

AI/ML engineer with 10+ years building production software and a deep specialization in non-deterministic AI systems: multi-agent orchestration, agent harnesses, retrieval and knowledge-base engineering, model optimization, LLM and RAG evaluation, and the high-concurrency Python services that put models in front of users. Specializations: AI Architecture, Data Science and Machine Learning, and Generative AI.

Ship agentic systems end to end — LangChain, LangGraph, and AutoGen orchestration; tool-use and skill design; vector and hybrid retrieval with re-ranking on Azure AI Search and AWS Bedrock; FastAPI services with async execution and streaming; and token- and latency-optimized inference paths that delivered 70% token reduction and 80% database load reduction on an executive AI platform. Built the USPS agent harness that improved SDLC workflow correctness and speed by 74%, and now lead engineering for the U.S. Department of Education's Aidan AI assistant.

Measure what ships with RAGAS and LLM-as-a-judge evaluation, and take ML from notebook to production with MLOps discipline (MLflow, Weights & Biases, LangSmith tracing) and treat safety as engineering: guardrails, red-teaming, bias mitigation, and PII redaction, from global retail scale to counter-disinformation NLP serving 10,000+ users. U.S. Air Force veteran; Founder of Vets Who Code; Microsoft Global Hackathon Winner, GitHub Star, and Google Developer Expert.

Technical Skills

Agentic AI & Orchestration: LangChain, LangGraph, AutoGen, LangSmith, multi-agent orchestration, agent harness design, custom skills and agent definitions, tool-use and function calling, MCP servers, context and knowledge-base engineering, GitHub Copilot extensibility, prompt engineering

LLMs & Foundation Models: OpenAI GPT-4o, Anthropic Claude, Google Gemini 2.5 Pro, Meta Llama, PHI-3, T5, Hugging Face Transformers, Transformers.js, Azure OpenAI, Amazon Bedrock

Model Training & Optimization: Fine-tuning and adapter methods (LoRA/QLoRA), DeepSpeed, deep learning, PyTorch, Scikit-learn, NLP, computer vision (OCR), speech-to-text, inference latency and token-cost optimization

Evaluation: RAGAS (faithfulness, answer relevancy, context precision and recall), LLM-as-a-judge, heuristic and rubric-based evals, LangSmith evaluations, golden datasets and regression test sets, red-team test suites

Retrieval & Vector Infrastructure: RAG, embeddings, vector search, hybrid search, re-ranking, Azure AI Search, Azure AI Document Intelligence, AWS Bedrock Knowledge Bases, PostgreSQL/pgvector, Redis

MLOps & LLMOps: MLflow, Weights & Biases, LangSmith tracing, model serving via REST/FastAPI, SageMaker, Docker, Kubernetes, GitHub Actions, Terraform, CI/CD, New Relic, structured logging, distributed tracing

Responsible AI & Safety: Guardrail design, red-teaming, adversarial testing, bias mitigation, Azure AI Content Safety (Prompt Shields), Presidio PII redaction, AI governance, generative AI data ethics

Data for ML: Python, Pandas, NumPy, SQL, data pipelines and migrations, Apache Kafka, Apache Spark, database performance management, star schema modeling, Power BI

Architecture & Backend: AI and ML architecture design, solution and cloud data architecture, security architecture, API design, FastAPI (async, streaming, concurrency), Pydantic, DBOS, Flask, Node.js, TypeScript, Java, event-driven design

Cloud & AI Interfaces: Azure (OpenAI, AI Search, AI Services, Speech, Functions), AWS (Bedrock, SageMaker, S3, SQS, Lambda), GCP, Firebase; Next.js, React, Angular, Streamlit, Tailwind CSS

Testing: Pytest, Jest, Playwright, k6 load testing, behavior-driven development (BDD)

Professional Experience

Accenture Federal Services

AI LLM Technology Architecture Manager (CL7) | April 2026 – Present | Remote

• Promoted to Engineering Manager for Aidan, the U.S. Department of Education's public-facing AI/RAG assistant used by Americans nationwide, leading a 14-person team as a player-coach while contributing to retrieval and model engineering (October 2026 – Present).
• As AI Enablement Lead on the United States Postal Service (USPS) program, built the agent harness around GitHub Copilot that completes work across the full SDLC, improving workflow correctness and speed by 74% and moving the tool from code suggestion to finished, reviewable work.
• Engineered every harness layer — custom skills, agent definitions, tool access, and a retrieval knowledge base spanning Angular, SQL, and Java codebases — so agent output is grounded in program architecture, standards, and domain context.
• Decomposed SDLC stages into scoped agent tasks with explicit tool access and handoffs, enabling multi-step agentic workflows to run to completion without a developer driving each step; shipped to production for the program delivery team.
• Built evaluation into the agent and RAG workflows with RAGAS (faithfulness, answer relevancy, context precision and recall) and LLM-as-a-judge scoring, gating releases on measured output quality.
• Set model selection, prompt design, and guardrail placement standards for agent-assisted delivery under Public Trust and federal compliance constraints.

Vista Equity Partners

Senior AI Engineer (Contract) | September 2025 – March 2026 | Remote

• Architected Q App, an agentic executive intelligence platform used by the CEO and leadership team to analyze $100B+ AUM across 90+ portfolio companies.
• Engineered the orchestration patterns behind Vista's Agentic AI Factory, the firm's strategy to deploy AI agents at scale across investment operations.
• Cut token usage 70% and database load 80% by implementing LangChain orchestration with Pydantic-typed schemas on Azure OpenAI and optimizing query and retrieval patterns.
• Evaluated Q App retrieval and agent output with RAGAS and LLM-as-a-judge evals over LangSmith traces, measuring faithfulness and context relevance as retrieval was optimized.
• Built Q API on FastAPI and DBOS with Azure OpenAI and Azure AI Search, delivering durable, asynchronous execution for long-running multi-step AI workflows.
• Exposed Salesforce CRM and Grata market intelligence as agent-callable tools, synthesizing heterogeneous investment data into executive-ready insight.
• Built generative AI pipelines that produce executive presentation decks and audio briefings, plus a Next.js/TypeScript front end for streaming, low-latency model output.

Serket-Tech Cyber Security

Senior Data Engineer (Contract) | September 2025 – December 2025 | Remote

• Designed the risk taxonomy and star schema data model behind an enterprise GRC Risk Register for a Fortune 500 healthcare company, integrating Application Security and Threat Management domains into one analytics-ready model.
• Delivered Power BI dashboards with real-time filtering and drill-through for executive and operational risk insight across multiple business units.

McGraw Hill

Senior AI Engineer (Contract) | September 2025 – November 2025 | Remote

• Architected AI Shepherd, a unified AI platform for scalable assistant deployment across educational products, and built AI agents with LangChain and AutoGen.
• Designed reusable integration patterns for AI workflows, connecting assistants to existing systems and reducing redundant builds; extended the Content Analysis Tool with a new question type.

Democratic Party

Director of Counter Misinformation / Disinformation Programming | January 2025 – August 2025 | Remote

• Built and optimized NLP and machine learning pipelines in Python on TikTok and X/Twitter API data to detect narrative manipulation, improving intervention speed by ~60%.
• Deployed LLM-powered assistants and RAG systems on Gemini and LangChain serving 10,000+ users under Responsible AI controls.
• Built automated fact-checking and trend-detection tooling, using adversarial testing and red-team review to reduce harmful and off-policy model output.
• Managed a six-person team delivering data models, analytics pipelines, and AI policy guidance; led AI literacy workshops on safe LLM use.

Microsoft

Software Engineer II, AI Systems & Infrastructure | March 2021 – January 2025 | Remote

• Built a full-stack RAG chat application for Domino's Pizza in Python, FastAPI, React, and TypeScript on Azure OpenAI, vector search, and LangChain, with Azure Content Safety and Presidio for moderation and PII protection; evaluated answer quality with RAGAS faithfulness and answer-relevancy metrics.
• Built a real-time speech-to-text system on Azure OpenAI and Azure Speech Services for Starbucks, streaming transcription at global scale with PII redaction in the request path.
• Productionized Scikit-learn and PyTorch models from experimental notebooks into scalable, observable REST APIs, improving provisioning time by ~50% and hardening MLOps with MLflow and Weights & Biases.
• Delivered an OCR and machine translation computer vision system in Python and Flask, reducing multilingual order errors by ~45% at global scale.
• Established LLM guardrail patterns — profanity detection and moderation layered into Azure Content Safety pipelines — reused across client engagements.
• Led Infrastructure as Code with Terraform and GitHub Actions, automating provisioning and CI/CD for production services.
• Engineered React and Python BI dashboards for Starbucks, cutting retrieval time ~50% and supporting a $2B contract win; authored an AI-900 study guide that enabled 100+ engineers to certify within two quarters.

Earlier Experience

• Rocket Technologies, Technology Evangelist (May 2020 – February 2021) — led developer enablement through workshops, seminars, and internal toolkits on new tools and best practices.
• CBS Interactive, Frontend Developer (February 2018 – June 2020) — built Ember and React interfaces across CBS digital platforms; prototyped features in Node.js.
• HigherVisibility, Web Developer (February 2015 – May 2016) — built Ruby on Rails applications and custom WordPress themes for international clients.

Leadership

Vets Who Code

Founder & Executive Director / Full-Stack AI Engineer | May 2014 – Present | Remote

• Built an LLM-driven success prediction and personalization model in Python with GPT-4o and Gemini, increasing accelerator completion by ~40% and reducing early attrition.
• Built NLP-powered resume rewriting tools with Transformers.js and T5, contributing to 300+ veterans landing tech roles.
• Built and maintain the Vets Who Code platform on Next.js and TypeScript with a FastAPI and Python backend, plus Firebase and GCP engagement data pipelines on Apache Kafka and Spark.
• Founded and scaled an open-source 501(c)(3) accelerator training veterans for software careers, with alumni earning $20M+ in tech salaries at Microsoft, Google, Home Depot, and others.

Teaching & Publications

• Frontend Masters, Instructor (March 2020 – Present) — three courses on engineering career readiness, 100,000+ hours viewed.
• LinkedIn Learning, Instructor (January 2023 – Present) — "Getting Started with GitHub," training thousands of developers.
• "Junior to Senior: An Action Plan for Engineering Career Success" — GitHub ReadME (2023); "Teaching in Public with GitHub" — GitHub ReadME (2022); "Hello World" series — Stack Overflow Blog (2020); "Open Source Opens Doors for Vets" — O'Reilly Radar (2018).

Awards, Certifications & Service

• Microsoft Global Hackathon Winner (2024); Google I/O 2025 Gemini Builder Invitee; Google Developer Expert; GitHub Star; Twilio Champion; LinkedIn Top Voice; White House Demo Day Honoree.
• Microsoft Certified: Azure AI Fundamentals (AI-900); LinkedIn Learning certificates in Conversational AI, NLP, and Computer Vision on Azure.
• United States Air Force, Security Forces (September 2004 – September 2009) — domestic and overseas security and law-enforcement roles across Iraq, Afghanistan, and South Korea. Active Public Trust clearance.