D
Don Martirez
Tailored ApplicationDisney Entertainment & ESPN Product & Technology (DEEP&T) — Business Operations AI Engineer
AI Engineer • 4+ Years Experience • Glendale, CA / Hybrid

I build and maintain production GenAI applications — grounded conversational systems, agentic workflows, and observable pipelines.

Software engineer specializing in Python-based generative AI systems, conversational interfaces for enterprise operational data, and deterministic retrieval workflows. Hands-on experience developing LangGraph multi-agent DAGs, GraphRAG in Neo4j, MCP-style tool integrations, and full-lifecycle LLMOps with golden evaluation suites.

Python & FastAPILangGraph & Multi-Agent DAGsGraphRAG & Neo4jMCP & Tool CallingNL-to-SQL & sqlglot ASTVector Search & EmbeddingsLLMOps & Golden Set EvalsDocker & Cloud RunCursor & Claude Code
Section 01 — Role Alignment • DEEP&T AI Engineer

How My Background Maps to Disney DEEP&T

Proven experience building conversational AI platforms, grounding responses in enterprise source data, and maintaining resilient production GenAI services.

  1. 01
    Python GenAI Application EngineeringStrong Python, backend services, web applications & operational data connectivity
    4+ years shipping production software. Built and containerized multiple production applications (the3pmdaily.com, artstache.app, don-the-imaginator.com) using Python, FastAPI, and Docker. Implemented asynchronous APIs, rigid Pydantic schema validation, and middleware connecting web apps with backend microservices—including a dedicated FastAPI microservice for natural language to SQL queries over 45 database tables.
  2. 02
    Retrieval & Grounding WorkflowsSource-backed AI responses, embeddings, vector search, semantic chunking & metadata
    Engineered deterministic GraphRAG in Neo4j AuraDB with LangGraph conditional DAGs. Modeled Lowenfeld's stages (1947) as property graph triples enforcing sequence: (:Stage)-[:PRECEDES]->(:Stage). Enforced a fail-closed grounding gate (is_grounded_in_neo4j returns null if 0 nodes match or confidence < 0.75). Verified Bolt Cypher query latency at ~90.07ms median local with an end-to-end <800ms target SLA.
  3. 03
    Agentic Workflows & MCP-Style Tool CallingTool use, function calling, API actions, multi-step reasoning & protocol integrations
    Implemented custom tool-calling engines rendering generative client UI components in real time. Designed MCP-style structured endpoints allowing conversational agents to execute deterministic Python tools (DCF calculations in 3.63ms median with 92% token savings, appointment scheduling, vector search). Engineered an enterprise NL-to-SQL tool allowing non-technical business operators to query 45 database tables in plain English: translates natural language queries into read-only SQL protected by sqlglot AST guardrails and a dedicated SELECT-only database role (nl_sql_reader)—completely eliminating risk of data corruption or unauthorized modifications—before hydrating interactive <NlSqlIntelligence /> visual charts and CSV exports. Client-side Pocket-Synth traps 80% of common queries locally with $0 LLM cost.
  4. 04
    LLMOps, Observability & Boundary-First TestingLogging, monitoring, latency tracking, cost management, golden datasets & troubleshooting
    Curated 50-vector golden evaluation datasets with 101 automated assertions running in 170ms in CI/CD (<800ms SLA). Mutation testing verified invariant resilience by catching 12 out-of-order failures upon edge deletion. Built automated heartbeat monitors for database pool warmups. Heavy compute runs locally to optimize cloud infrastructure costs to ~$18/month.
  5. 05
    Modern Engineering Velocity & Cloud DeploymentCursor, Claude Code, Docker, CI/CD, and cloud environments (GCP production + Azure concepts)
    Required daily use of AI coding tools (Cursor, Claude Code, Gemini CLI) for high-velocity full-stack engineering, test automation, and documentation. Deployed containerized microservices to Google Cloud Run with automated CI/CD; familiar with Azure OpenAI, Azure Functions, and App Service architectures.
Section 02 — Production Case Studies

Production GenAI Systems

Conversational Platforms • Grounding • LLMOps
Closing Terminal • Python & TA-LibLangGraph • Cloud Run • Live

3PM Daily: Solo-Operated Market Closing Terminal

Problem

Operators and traders drown in 800+ tickers during the chaotic 2-hour market close, causing analysis paralysis and missed setups.

Engineering Solution

Built a two-stage closing pipeline that strictly separates compute from interpretation: C-compiled TA-Lib deterministically calculates mathematical stops across ~827 symbols in 4.2s, while a 9-agent LangGraph debate evaluates fundamental risk and filters down to 4 high-conviction setups.

Verified Outcome

~$18/month total infrastructure cost with >90% gross margin. 101 CI/CD regression assertions pass in 170ms with zero hallucinated prices reaching client cockpits.

GraphRAG • Source Grounding • LangGraphNeo4j • Python • Live

Artstache: Source-Backed Pedagogical GraphRAG System

Problem

Parents reviewing children's artwork at 6:15 PM face paper clutter and fatigue. Vector search conflates a 4yo tadpole with a 10yo comic, risking out-of-sequence pedagogical recommendations.

Engineering Solution

Modeled developmental stages as Neo4j property graphs (:Stage)-[:PRECEDES]->(:Stage). A 5-node LangGraph DAG only synthesizes guidance grounded in verified graph nodes, while an in-memory Canvas pipeline strips all EXIF GPS metadata before upload.

Verified Outcome

~90.07ms median Cypher traversal with 50/50 golden vectors passing. Mutation testing (deleting a graph edge) empirically verified the fail-closed gate by catching 12 out-of-order attempts.

Conversational Platform • Tool CallingFastAPI • Python Tools • LiveNL-to-SQL • 45 Tables

don-the-imaginator.com: From Gated Friction to Open IA

Problem

Mandatory Firebase auth gate created severe drop-off: Firestore SiteAnalytics revealed Funnel: 20% signup, 0% activation (N~10/mo). Exclusivity hypothesis failed; technical evaluators demand immediate proof without friction.

Engineering Solution

Bifurcated the architecture: a reserved landing for rapid 30-second reviews and an isolated /lab for live technical experiments. Deployed client-side query trapping (80% resolved at $0 LLM cost) alongside an enterprise NL-to-SQL engine enabling non-technical operators to query 45 database tables with read-only AST safety rails.

Verified Outcome

Immediate time-to-proof (<5s) with 0.00 median CLS. Enterprise data safety guaranteed via AST single-statement validation and physical SELECT-only read permissions over all 45 tables.

Section 03 — Engineering Standards & LLMOps

How I Build Resilient GenAI Systems

Rigorous production standards from schema design and deterministic grounding to continuous observability and regression testing.

  • 01
    Standard 01

    Python Microservices & API Integration

    FastAPI services with asynchronous request handling, Pydantic schemas, and structured error boundaries connecting web interfaces to external data stores.

  • 02
    Standard 02

    Deterministic Grounding & GraphRAG

    Property graph modeling in Neo4j and vector search indexers that enforce source-backed responses and eliminate hallucinated sequences via [:PRECEDES] constraints.

  • 03
    Standard 03

    Agentic Workflows & MCP Tool Calling

    LangGraph orchestration DAGs with human-in-the-loop validation, multi-step tool execution, and real-time generative UI component rendering.

  • 04
    Standard 04

    Full-Lifecycle LLMOps & Observability

    Automated telemetry, structured JSON logging, token cost tracking, and heartbeat monitors maintaining database pool availability.

  • 05
    Standard 05

    Golden Datasets & Boundary-First Testing

    Pre-deployment validation suites using 50-prompt golden datasets, 101 automated assertions in 170ms in CI/CD, and mutation testing before production release.

  • 06
    Standard 06

    Containerization & Developer Velocity

    Docker containerization and Cloud Run deployments accelerated by modern AI developer workflows (Cursor, Claude Code, Gemini CLI).

Dedicated to enterprise AI safety, access control, and source provenanceCollaborative engineering with senior teams • Clear technical documentation & retrospectives