Andrew
DiCosmo

A live AI-enabled marketplace for collectors. I designed and built a stateful LangGraph identification workflow coordinating computer vision, catalog matching, licensed-reference retrieval, web and marketplace research, confidence validation, follow-up questions, retries, parallel execution, pause and resume, and human approval. LangChain standardizes model, tool, retriever, and middleware integrations; LlamaIndex retrieves across catalogs, scanned guides, manuals, checklists, and pricing references. Amazon Bedrock AgentCore supports governed browser automation, identity, policy enforcement, memory, and observability. CrewAI and AutoGen are under evaluation for future specialist-agent workflows.

A private Shopify repricing system where an LLM recommends prices, but deterministic policy and validation code control every write. The model gets an opinion; the guardrails get the final say.

An in-development business-operations platform. The current work focuses on shared customer and hiring workflows, with AI-assisted matching, drafting, and planning. Additional modules are on the roadmap. See the concept view below.

After BlackSky ended in March 2023, Citizens Bank became my primary role. The work began through a company-to-company engagement and later converted to W-2, effectively a contract-to-hire progression. I served as the Azure security subject matter expert for the content engineering team, owned the Azure security product workstream, and supported adoption of Azure Advanced Threat Protection, Office 365 Advanced Threat Protection, and the broader Microsoft security suite. I also took on QRadar work outside the original Azure scope, helped the team write Splunk detection rules during the migration from QRadar, and built an early Detection as Code prototype that used source control for change history, recovery, and rule restoration. I automated detection and SOAR playbooks within Citizens Bank Cyber Defense, using Wiz to identify high-severity findings that activated event triggers in Splunk. Later work included Microsoft Purview data flowing into Splunk.

Current C2C subcontractor engagement supporting FDIC as the end customer across more than 60 Azure subscriptions. I engineered hundreds of Azure Policy and policy-as-code controls aligned with the agency's customized Azure Security Benchmark and NIST requirements, then automated evidence, drift detection, findings, and remediation with PowerShell, Wiz, Defender for Cloud, and GitHub Actions. The workflow reduced compliance investigations that could take most of a business day to under a minute. I also use Azure Resource Graph, Azure Monitor, Log Analytics, and Splunk SPL for telemetry validation, dashboards, correlation searches, alerting, and security monitoring, and built automation to identify unsupported Azure App Service runtimes and notify resource owners.

C2C subcontractor engagements supporting Fortive, HCSC, and Guidestone as end customers: deployed and configured Microsoft Defender for Endpoint and Defender for Identity, reducing attack surface across corporate environments. Led Azure ATP implementations for real time threat detection, integrated Splunk and QRadar with the Defender stack across hybrid environments, and built Defender policy and automated security responses across the three end clients.

During my direct BlackSpoke role from 2017 to 2021, I led cloud architecture and delivery for DOJ and NGA programs, migrated Justice Management Division datacenter workloads into Azure, and helped the NGA team maintain and extend the Corporate Analytic Visualization Environment after the prime contract win. I also led two teams building mission microservices and personally developed Python and Java Spring APIs that cataloged petabytes of satellite imagery, along with data pipelines processing terabytes daily from multiple sources. The container platform moved from Kubernetes on EC2 to AWS EKS. In a separate later CloudTect subcontract engagement supporting NGA through BlackSpoke, we moved a web, API, and database microservice to OpenShift using Docker images, Helm, private registries, and GitLab CI/CD. Terraform, Prometheus, Grafana, Kibana, Elasticsearch, Lambda, NiFi, SQL, Oracle, and SSIS supported infrastructure, observability, and data delivery.

At BlackSky Technology (NYSE: BKSY), worked on Spectra, an AI-enabled space-intelligence platform that fused satellite imagery with SAR, RF, GPS, and IoT sensor data for object, change, and anomaly detection and site monitoring. The work included PyTorch and TensorFlow detection models, a continued NGA AWS microservices contract, and an IARPA program. Platform work used EKS-based microservices, Docker, Helm, Prometheus, Grafana, Kibana/Elasticsearch, and Airflow to monitor container performance, tune CPU utilization, and support scaling decisions. The visual on this site is an original illustration, not client material.

The largest production AI application I have worked on was a multi-year modernization of an existing cleared mission application. I led two delivery teams rearchitecting services on AWS using Well-Architected principles and loosely coupled microservices behind APIs. My teams owned Python API services end to end; I later worked alongside Java backend services in the same environment. The program operated at petabyte scale, with data warehousing measured in billions of records and continuous structured and unstructured ingest supporting mission-focused AI/ML, computer-vision, and geospatial workflows.

At HARMAN, I led the architecture and application design for the original Sherlock AI product and worked hands-on with two other talented developers to build it as a rapid prototype. Sherlock AI combined data scraping, machine learning, and predictive analytics to analyze market signals for build-versus-buy decisions and guide R&D investment. The platform continued in internal use after my departure. Built on Microsoft cloud services, Sherlock AI received a 2016 Digital Leader Award from IDG and Dimension Data.

C2C subcontractor engagement supporting NIKE as the end customer. Built a two-year technology roadmap for NIKE's corporate information security team covering application security, cloud platform security, encryption and certificate management, vulnerability management, and penetration testing, and integrated CloudWatch, Rapid7, and AlienVault across 40+ Azure and AWS subscriptions.

As ByteCubed's Principal Big Data Architect, led an engineering team building big data platform solutions at scale, with DHS among the customers served. Owned software architectures, coding standards, code reviews, and the technical direction that kept feature delivery on time.

Eleven years teaching nights while building days: Visual Basic, Java, C#, C++, and Ethical Hacking, plus a seat on the IT advisory board. Also served as an exam development Subject Matter Expert for CompTIA and Microsoft. Explaining systems is how you find out if you understand them.

Directed IT for eight affiliated companies and led ERP and messaging migrations for 300 users. Built an inventory-tracking system for overseas factories that produced barcode labels used to manage warehouse logistics as products were repackaged for delivery to big-box retailers.

My own shop: custom desktop, web, and mobile apps for small and medium businesses, SharePoint and ASP.NET builds, network design, and cloud work that was early for its time: refactoring client applications to run in AWS and Azure and standing up DevOps pipelines for them. I led client development, scoping, delivery, and support while managing developers and subcontractors across software, web, ERP integration, and LAN/WAN projects.

Handled the complete internal technology function, from break-fix support and infrastructure through ERP workflows and custom software. I built a validated material-record system and connected it to ERP shipping output, eliminating a tedious manual printing step and saving approximately 80 to 125 hours per year. I also automated plant-manager job-cost review from roughly one hour per job to a button-driven process across approximately 15 to 30 completed jobs per day, making profitability problems and data-entry errors visible much sooner. After leaving as a W-2 employee, I continued providing part-time and occasional onsite support for roughly 15 years, then helped train and transition responsibility to the internal IT team.