Adam Albright¶

Titusville, FL | 321-837-9952 | aalbright@iamrehket.me | LinkedIn | GitHub

Summary¶

Software and platform engineer with more than eight years at Office Depot / The ODP Corporation (now ODP Group), a Fortune 500 retailer, designing, building, and operating enterprise systems end to end: governed data-access platforms with fine-grained authorization, search and matching engines over 170,000-product catalogs, cross-cloud (AWS to Azure) service migrations, public-facing hardened services, and high-volume data pipelines that are still in production years after being handed off. Technical lead for a five-engineer AI platform team since 2025; hands-on manager of 8-13 engineers and QA for three years before that. Author of engineering standards adopted across other teams' services: CI quality gates, a payload-validation pattern picked up by more than five services, base container images, infrastructure-template validation. Adviser to executives at affiliate companies on AI adoption and governance.

The recurring theme across every project: the problem was rarely the algorithm. It was data access, data ownership, and building something that keeps working after the team moves on.

Professional Experience¶

ODP Group (Office Depot) - Sr. Lead Software Engineer, AI Platform¶

June 2026 - Present · Boca Raton, FL (remote)

Returned to the individual-contributor track after the interim management assignment below. Technical lead for five engineers.

Lead engineer for an enterprise data catalog and governed data-access platform I designed earlier in 2026. I own the architecture, the roadmap, and the standards the team builds against, and I review every significant change.

Implementation and rollout of the platform: modular adapters connecting upstream data sources; artifacts binding an adapter to a specific table or data element; per-team data groups where each consuming team owns its own taxonomy and definitions; and separate MCP, HTTP, and SDK interfaces into a shared core so agents, REST callers, the web UI, and programmatic consumers each get an appropriate surface.
Authorization model in production: upstream credentials terminate at the platform, and per-user access is re-derived from platform roles evaluated by OpenFGA running as a sidecar against Postgres. Entra integration supports three authentication paths - directly against the API, through the web application, and through an agent such as Claude or Copilot. SCIM drives control-plane management (roles, groups, tenants, users), with auditing of both control-plane changes and data-access events.
Onboarded the pricing organization, which now works with governed company data through Claude over MCP.
Deployment properties: multi-tenant; pluggable secrets subsystem (Azure Key Vault in production, extensible to other stores without core changes); fully offline/air-gapped operation from Docker artifacts with either the customer's identity provider or a bundled Authentik instance. Delivered the team's first fully automated end-to-end CI/CD test suite, now the reference pattern for other services.
Built the sensitive-data scanning and classification system for the Snowflake warehouse at the security architecture team's request (see Selected Engineering Problems); the platform consumes its audit record to gate dataset onboarding.
Built and maintain guardrails for AI-assisted development: synchronized agent instructions, static analysis, and checks for unintended growth in the application's taxonomy, each added in response to an observed coding-agent failure. Established a rule that every CI check must have a locally runnable equivalent, so developers can catch failures during development instead of waiting for a 15-minute CI run.

ODP Group - Manager, Sr. Technology - AI Transformation (interim)¶

January 2026 - June 2026 · Boca Raton, FL (remote)

Interim management assignment during a reorganization, with continued hands-on engineering responsibility.

Designed the enterprise data-access platform - requirements, architecture, component model, authorization model, and roadmap - after two triggers: a forward-deployed engagement where my team was starting from scratch while another team prototyped on a platform, and the question that has stalled nearly every project I have worked on here: who owns this data, who manages it, where is it, and how do I get access. Made the case for a modular adapter architecture after showing that a single integration API could not accommodate the number of sources or the different workflows individual teams require; reusable source adapters were a new pattern for the organization.
Built the product-alternates search service: proposed Azure AI Search as the approach, designed the pipeline consolidating Snowflake, SharePoint, and Postgres into a maintained index, and built the FastAPI service the system runs on. Set a pass/fail criterion: the correct alternate had to appear in the top three results. Evaluated cases using a known-product test harness reviewed with merchandising subject-matter experts, and tuned the search configuration against it. Two engineers built on that framework: one integrating the existing product-search service, one on prompt refinement and final classification.
Led the demand-forecasting work for a forward-deployed engagement at a portfolio manufacturing company: ran an AutoML-style workflow iterating models and ensembles for demand prediction alongside teammates' profit-based and time-series-LLM forecasts. Input was roughly three years of periodic canned reports rather than order-level history. Aggregate forecasts looked reasonable, then fell apart when filtered by product category or product type; intermittent-demand products were not modelable with the available data. Reported an honest no-go, mapped the manufacturing order lifecycle with site operations as the prerequisite artifact, and wrote the pre-engagement data-access, data-quality, and data-requirements recommendations for future deployments.
Managed the AI platform team: open-ended 1:1s, working sessions on CI/CD, testing, and containerized end-to-end flows, recorded into the team knowledge base. Joined the IT organization's Technical Leadership Community; planned and organized its recurring knowledge-sharing sessions, led the April sessions, and delivered presentations on AI tooling, skills, software development practices, and security.
Advised executives at multiple portfolio companies on AI adoption, governance, implementation strategy, and platform considerations.
Maintained deployments and production support for the team's services through the first quarter, and planned the wind-down of the team's earlier AI chat platform as the governed data-access platform became its successor.

The ODP Corporation - Sr. Lead Software Engineer¶

April 2025 - January 2026 · Boca Raton, FL (remote)

Technical lead for five engineers across concurrent AI data-integration projects.

Led design of AI data assistants for sales transaction history and inventory analysis: data catalogs, metadata, and taxonomies built with subject-matter experts so an LLM assistant could generate correct SQL for non-technical users. Built checks on generated-query correctness and guardrails that tell users when a question leaves covered scope.
Established the validation method - golden question/query pairs plus SME review cycles - and used it to find the real problem: generated queries were wrong roughly 75% of the time because the reference tables supplied by the first user group were not adequate for organization-wide use. Rebuilt the reference metadata with the BI team; correctness rose to roughly 80% in our evaluation tests.
Discovered that two consumer groups used identical terms with conflicting meanings (vendor, dropship, externally facilitated orders), so tuning for one measurably degraded the other. Resolved it with a separately governed vocabulary per group - the design that later became the data-access platform's data groups.
Architected the product-alternates engine over roughly 170,000 products: request path embeds the input product, retrieves candidates, and re-ranks with an LLM prompt set. Evaluated cosine similarity and BM25, then selected hybrid retrieval combining BM25 keyword search with vector search. Tested Postgres index types and selected Azure AI Search for the index, with raw product data in Postgres and a hosted OpenAI model for embeddings. Replaced the finance taxonomy that had produced cross-category false matches in the earlier system with merchandising's consumer-product taxonomy and filtered candidates to subclass/class before matching.
Defined the engine's data-lifecycle standard: idempotent Kubernetes sync jobs against live sources, restartable by re-trigger or from a named step; pre-flight and post-flight validation on every run; automated tests on the jobs and the service; and a curation tool giving merchandising visibility into why an item matched (for example, supplier misclassification) so they can fix source data. Their feedback tuned the re-ranking prompts. Delegated per-source retrieval and reconciliation to team members against that framework. In production for merchandising operations and a sales workflow where reps upload competitor product lists and receive equivalent company inventory.
Designed service-account isolation and schema-level permissions to restrict LLM-generated queries to authorized datasets. Built training datasets and table definitions that became reusable patterns for later data-source integrations.
Led incident response and outage calls; drove adoption of Bicep template validation and a Python/FastAPI base image.

The ODP Corporation - Manager, Technology¶

April 2022 - April 2025 · Boca Raton, FL (remote)

Hands-on hybrid role: daily design and implementation work, with performance management and long-term planning for 8-13 developers and QA engineers across the e-commerce and CRM platforms.

Cross-cloud migration program (AWS EKS and Tomcat VMs to Azure AKS) across Node/React micro-apps, Java Spring Boot, Go/Gin, and Python services. The bulk of the work was not the cluster itself: I built the tooling to create and update Docker images and repositories, produced the service inventory and definitions (no one had a complete list of what ran or what it was for - the gap that drove the API Management evaluation), supported each team's cutover plan, and wrote the procedures teams used to run and performance-test services whose databases stayed in AWS while the service moved. Built hardened, public-facing proxy services bridging the two clouds mid-migration, with rules across the web application firewall and payload-verification libraries in Go, plus test suites and documentation. The program was later curtailed on cost and shifting priorities; the tooling, hardened services, and procedures remained in use.
Delivered a public communication-preferences service (case study), adapting request validation and edge protections to changing infrastructure constraints; the validation pattern was adopted by more than five services.
Credential-stuffing detection on login endpoints (case study): detected while on call on systems another team owned, escalated through Security Operations, built the compensating detection control.
Led login modernization across the company's e-commerce platform; the rollout of a second authentication factor on the e-commerce web login pages mitigated credential-stuffing attacks against those pages.
Release management: monthly enterprise releases across roughly 40 services plus the legacy Tomcat codebase serving retail, business, and internal sites; owned requirements, timeline, dependency coordination, code freeze, and Saturday early-morning production deployments; real-time triage and rollback decisions. Built Python automation validating code commits against Jira tracking to find release gaps before deployment day, and Azure OpenAI tooling that scanned Jira and GitHub changes across services to produce release summaries and delivery catalogs - a multi-day manual task reduced to a script, used for about a year across the roughly six releases I ran and by successors.
Incident response: roughly five major incidents per year; pulled into cross-organization incident calls when they stalled; traced root causes across e-commerce services and ran a condensed release process for emergency fixes.
Implemented LLM classification of NPS customer feedback (order issue vs. customer service vs. payment) with Azure OpenAI, including the pipeline that landed and prepared the data.
Standardized Bicep template validation and a Python/FastAPI base image; built Node Docker images for registry compatibility; integrated Dynatrace into Docker images ahead of peak sales events; retired unused services; ran technical sessions and demos supporting broader Azure adoption.
Programs: business-unit separation; CRM and marketing technology stack separation between Office Depot and ODP Business; Enterprise Territory Management rollout; Anaplan compensation platform rollout; AI chat platform integration into Salesforce and the e-commerce API (2024).
Hiring, performance management, and mentoring; developed junior contractors into strong backend and CRM developers; established technical hiring standards after early hiring problems.

Office Depot / The ODP Corporation - Senior Software Engineer¶

September 2020 - April 2022 · Boca Raton, FL (remote)

Designed and built a CRM-to-warehouse sync framework (case study): declarative, template-defined bidirectional synchronization between Salesforce objects and Snowflake that replaced IBM DataStage jobs which timed out arbitrarily and paged on-call essentially every night. Roughly 300 GB nightly, covering most of the Salesforce platform's data; 20-million-record contact and account objects that previously overran the next day's scheduled run now complete in hours; pages went from nightly to a couple per month, then lower after per-batch retry with exponential backoff. Still in production in 2026.
Set up the cutover process and handed migration of remaining objects to team members and contractors after the libraries and main objects were in place.
Built reconciliation workflows for the harder direction (Snowflake into Salesforce), avoiding duplicate records and records stuck on partial payloads against systems Salesforce itself synced with, such as order management.
Data engineering between Salesforce and Anaplan for territory assignment (Enterprise Territory Management): Anaplan had hard volume limits, so everything was trimmed and orchestrated upstream - the sync framework into Snowflake, refined views of current territory assignment and customer volume, then Anaplan.
Deployment automation for customer data capture, cleansing, deduplication, D&B enrichment, rep assignment, hierarchy management, and compensation impact; integrations with Snowflake, the finance system, the compensation platform, and order management. Custom rule engine and data-sync framework between Salesforce and Snowflake. Ansible/AWX adoption for data consistency. Salesforce org modernization and dead-code removal. Daily Salesforce office hours. 100+ stories and bugs delivered in 2021.
Projects led: self-service account creation; sales-rep deployment automation; sales-play campaign integration to Salesforce Marketing Cloud; quote-to-order with Etilize content enrichment; pricing conversion.

Office Depot - Software Engineer¶

March 2019 - September 2020 · Boca Raton, FL

Architected and built the CRM platform's product-search service API (FastAPI, PostgreSQL, loaded daily from S3), which served the platform's SKU quoting capability and became the template for the team's Python applications.
Designed and implemented a record-linkage service (Django REST Framework) that consolidated the customer relationship hierarchy and prevented duplicate record creation in sales workflows.
Built an earlier product-search index letting sales reps find product alternates by cosine similarity over product descriptions. It drove real customer sales and was hard to maintain - very similar descriptions produced bad matches under a finance-oriented taxonomy, and there was no data lifecycle. Those failures directly informed the later product-alternates engine.
Built a request-dispatch framework syncing CRM support cases and payloads to downstream applications with FIFO priority queuing, and ETL tooling on Ansible and Ansible Tower with custom Python CLI tools.
Led the transition from legacy HTML templating to Lightning Web Components with admin-configurable components that removed developer dependencies; built a configurable, reusable lead-capture UI.
Instituted team training on DRY, postmortems, secure input handling, and functional programming.

Office Depot - Software Engineer, Associate¶

January 2018 - March 2019 · Boca Raton, FL

Sole designer and implementer of customer classification and decision-maker identification for a customer-data program, handed to the CRM team with no prior organizational expertise in this kind of system. Entity resolution across roughly 18 million contacts, 2 million accounts, and 21 million ship-to records pulled from Salesforce (Python, scikit-learn, fuzzy matching); k-means clustering mapped to spend-potential tiers; an entry-time duplicate-prevention service that searched existing customers, ship-tos, and contacts before creating new records. Built the Salesforce mass-export tooling that did not exist.
Outcome and lesson: a one-shot cleanup with no ongoing pipeline meant the dataset drifted immediately after go-live; without a way to trace a tier back to its inputs, sales reps lost trust in the classifications, and the model was replaced by a rules-based workflow within weeks. Decision-maker data collected under three teams' different definitions could not be used coherently. Everything I have built since on data catalogs, traceability, and post-deployment governance comes from this project.
Tied for first place in the inaugural Office Depot hackathon: a PyTorch image-processing pipeline over demo-store security-camera feeds, mapping camera regions to store zones and pixel-change thresholds to occupancy, producing heat maps of customer presence over time.

Education¶

University of Central Florida, Orlando, FL - Bachelor of Science, Computer Science, December 2017.