Diego S.

I work as a forward-deployed AI engineer — embedded directly with your team to design, build, and ship production AI systems, not just advise on them. Most AI engineers can wire up an API; few understand data at depth. I do — and that combination is the whole point. My background spans 20+ years at PricewaterhouseCoopers (PwC) and Publicis Sapient — seventeen years building software, data, and machine-learning systems, and the last three focused entirely on generative AI. That lets me bridge business strategy and technical execution for Fortune 500s and growth-stage companies. Clients span ADT, Aflac, AT&T, Capital One, Carnival, Lumen (formerly CenturyLink), Eli Lilly, Fanatics, HBO, IBM, Lennar, NBCUniversal, USAA, Visa, Wells Fargo, and Worldpay. Degrees in Computer Science and Applied Statistics, plus an MBA. 𝗪𝗵𝗮𝘁 𝗜 𝗯𝘂𝗶𝗹𝗱 Agentic AI systems — autonomous agents that reason, make multi-step decisions, and take real-world action through tools and APIs. I own the full design: LLM selection, RAG architecture, MCP server development, multi-agent orchestration, and production deployment on AWS, Azure, or GCP. The data-science layer comes built in — pipelines, predictive models, and evaluation — so agent decisions are trustworthy and measurable, not black boxes. 𝗪𝗵𝘆 𝗺𝘆 𝗯𝗮𝗰𝗸𝗴𝗿𝗼𝘂𝗻𝗱 𝗶𝘀 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 Most AI engineers don't understand data at depth. My foundation spans the full data-science stack — statistical inference, supervised and unsupervised machine learning, time-series, propensity scoring, customer segmentation, and marketing-mix and incrementality modeling — across financial services and fintech, insurance, telecom, healthcare and life sciences, media and entertainment, retail and e-commerce, consumer goods, and hospitality and travel. That depth changes how I build AI: models that are properly validated, pipelines designed for data quality, and outputs judged against real business metrics — not benchmark scores. 𝗦𝗲𝗹𝗲𝗰𝘁𝗲𝗱 𝘄𝗼𝗿𝗸 — 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜 • ContactOS — a multi-tenant AI customer-service platform in production across voice (Twilio, Deepgram, ElevenLabs), WhatsApp, web chat, and outbound campaigns. LangGraph ReAct agents call live customer APIs in real time; per-tenant configuration via a runtime registry (a new client is a new directory, not a code change); human escalation over a WebSocket dashboard. • Replaced a specialty retailer's HubSpot CRM with a production FastAPI + React platform in weeks — built with agentic AI development tooling throughout, with AI email summarization and a shared-database architecture with the client's analytics portal, live at a fraction of the prior license cost. • Two-phase agentic engineering triage — a fast LLM classifier routes inbound tickets in under 10 seconds, then an investigation agent autonomously traces the codebase and posts structured root-cause findings to Jira. • End-to-end RAG content engine — ingests YouTube transcripts, PDFs, Kindle highlights, and web articles, then generates channel-specific content grounded in the author's own voice (not generic AI output), auto-publishing to WordPress with a companion audio file. 𝗧𝗵𝗲 𝗱𝗮𝘁𝗮 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝘂𝗻𝗱𝗲𝗿𝗻𝗲𝗮𝘁𝗵 The same modeling rigor I bring to making agent decisions measurable: • Telecom propensity — scored 6.1M customers across 200+ behavioral indicators into calibrated propensity tiers, prioritizing the highest-likelihood segments to steer self-service adoption and call deflection. • Supplemental insurer media-mix — modeled a $171M media portfolio with adstock and saturation, producing per-channel ROI with credibility intervals to guide budget reallocation. • CPG segmentation — an RFM + cohort-LTV engine on Databricks classifying 83K customers weekly, driving segment-level targeting across discounting, channel, and reactivation. 𝗖𝗼𝗿𝗲 𝘀𝘁𝗮𝗰𝗸 Python · SQL · R · PySpark | LangGraph · LangChain · CrewAI · n8n · MCP Servers | Anthropic Claude · OpenAI · Ollama | Pinecone · Qdrant · ChromaDB · pgvector · mem0 | Twilio · Deepgram · ElevenLabs · Whisper | FastAPI · React · TypeScript · Docker · GitHub Actions | PostgreSQL · MongoDB · Snowflake · Redshift | dbt · Databricks | XGBoost · scikit-learn · MLflow | Tableau · Power BI · ECharts | AWS · Azure · GCP 𝗛𝗼𝘄 𝗜 𝘄𝗼𝗿𝗸 I take ownership of delivery — I scope, architect, build, and deploy, and communicate clearly with technical and business stakeholders throughout. Whether you need a solo IC or a lead to drive a small team, the approach is the same: embedded, accountable, and shipping.