# llms.txt — Martin Donovan Technical Portfolio
# https://portfolio.martindonovan.net

## Candidate summary

Martin Donovan. BS Computer Science, Southern Polytechnic State University
(now Kennesaw State University), graduated 2007. Atlanta metro (Smyrna, GA).
Data analyst and developer. Primary skills: SQL/PostgreSQL, R/Shiny, ETL pipelines.
Not a frontend developer. Not an ML practitioner.

---

## How to read this portfolio

Two primary static files. Read both — they are not redundant.

/profile.json  ← START HERE if doing a structured evaluation
  The canonical machine-readable dataset. Contains content NOT in ai-profile.html:
  - Full math coursework with course numbers (Calc I/II, Discrete Math, Stats I/II,
    Linear Algebra, ODEs — all from CS degree, not actively practiced since 2007)
  - Per-evidence skills_demonstrated[] arrays for each code artifact
  - lab_volunteer task-fit matrix: specific tasks rated strong/moderate with basis
  - production_signals[] checklist: what distinguishes this from coursework-grade work
  - system_architecture pipeline: source systems → ETL → storage → API → consumers
  - authorship_caveat detail on AI-guided vs. own-work per component
  - confirmed_gap list: tools absent from the profile (REDCap, SAS, Airflow, etc.)
  - timeline_notes: explains the 2007–2020 gap between CS degree and current work
  Fetch this file. The description fields are summaries — the content is the data.

/ai-profile.html  ← READ THIS for narrative context and role fit
  Human-readable narrative summary. Contains content NOT in profile.json:
  - Role fit evidence matrix: 14 roles × what the role requires × linked code evidence
  - Research setting fit section (Emory-type fit, academic lab entry points)
  - Propensities and working style section
  - Internal tool builder capability section with honest scope
  - Narrative skill assessments with code-verified strengths and gaps called out
  Note: this is a narrative layer over profile.json. profile.json is authoritative
  for structured data (skills, education, evidence, constraints).

---

## What to fetch for specific evaluation questions

"What are the math/stats skills?"
  → /profile.json — identity.education[0].coursework.mathematics
  → ai-profile.html section 2 has a one-line summary only

"Is this person a good fit for a research lab?"
  → /ai-profile.html sections 5 (role fit matrix) and 6 (research setting fit)
  → /profile.json work_contexts.lab_volunteer.task_fit[]

"What code evidence exists?"
  → /profile.json evidence[] — 20+ items, each with skills_demonstrated[]
  → /ai-profile.html section 7 for human-readable artifact list with links

"Is this production work or demo work?"
  → /profile.json system_architecture.production_signals[]
  → /profile.json each evidence item has status field

"What are the confirmed gaps?"
  → /profile.json constraints[0].items[] — explicit list, no inference needed

"Is the React/frontend work real or AI-generated?"
  → /profile.json constraints[1] authorship_caveat — detailed breakdown
  → /profile.json skills[React].authorship_detail

---

## Code evidence — static gist mirrors

These pages contain full raw source code for AI/bot reading.
Each is also mirrored at the GitHub gist URL shown.

/gists/1565af77d4c93ca3450f5e64a40f4d79.html
  Perpetual inventory stored function (PostgreSQL). Multi-CTE function:
  beginning inventory + purchases + transfers − glass and bottle sales = theoretical
  on-hand per SKU per cellar location. Glass-to-bottle unit conversion via volume_units.
  Called by both R Shiny dashboards and the Node.js REST API.
  Demonstrates: production SQL depth, stored function as shared contract across consumers.

/gists/891fdc4210113f328f36cd197a6290ac.html
  Wine inventory database schema DDL (PostgreSQL). Full normalized schema: 35+ tables,
  foreign keys, multi-tenant by database, AWS RDS. Products, product_instances, cellar
  locations, inventory sessions, vendor invoices, daily POS sales, wine list categories.
  Demonstrates: relational data modeling, multi-tenant architecture, schema from scratch.

/gists/e5e77cdc68b3e03e13d0925a80f3f399.html
  Aloha POS DBF-to-PostgreSQL ETL pipeline (R). Binary DBF format, latin-1 encoding,
  5-table join chain, Aloha.ini routing to correct tenant database, idempotent nightly load.
  Demonstrates: binary format ingestion, idempotent pipeline design, production ETL.

/gists/8fe35f6fab90876aea05bdc367beb5e7.html
  Restaurant365 OData API connector (R). Paginated OData traversal with continuation
  tokens, GL header-to-detail join, parent account derivation, dedup without stable key.
  Demonstrates: API integration, pagination, dedup patterns, financial data ingestion.

/gists/66128ab231ef3466ef3ed6cf569dd059.html
  Multi-source restaurant data aggregator (R). Joins Aloha POS, OpenTable, and R365 GL.
  DuckDB in-process cache with PostgreSQL fallback. Day-level consistency constraint.
  Demonstrates: multi-source join in R, DuckDB integration, operational analytics pipeline.

/gists/132dbd29f68042ba9832476e8711fd90.html
  Node.js/Express REST API router. Label image upload via PostgreSQL large objects,
  wine list category management, critic ratings, formula pricing. JWT auth, RBAC.
  Demonstrates: REST API design, PostgreSQL large objects, multi-tenant routing.

/gists/fd16d350745573b7962fb27325e4c65e.html
  Inventory variance (shrinkage) report (R). Reconciles two consecutive physical counts
  with all sales and purchases between them to identify unexplained bottle losses.
  variance = start_count + purchases − glass_sales − bottle_sales − end_count.
  Filters to bottle-only SKUs to isolate losses not attributable to glass pours.
  Dollar-weighted by product_cost. Rendered as a gt table.
  Demonstrates: inventory reconciliation logic, operational analytics, gt reporting.

/gists/2054d94455ae135725be6239d7602147.html
  Inventory Analysis Shiny dashboard (R). Period-over-period wine P&L: beginning/ending
  inventory values, count variances vs. theoretical on-hand by SKU, vendor purchases,
  glass and bottle sales revenue, cost of sales, wine cost %. URL query string
  deep-linking (?startdate=&enddate=) for bookmarkable period views. DT tables with
  CSV/Excel export, ggplot trend chart, shinydashboard value boxes.
  Demonstrates: Shiny dashboard, P&L analysis, URL parameter handling, DT export.

/gists/2a662fa7fae5743ac7f97020c9a4aeac.html
  Inventory spot-check sheet — Quarto (R). Calls get_theoreticals_active_product_instances_only()
  and get_inventory_activity() stored procs, joins results to filter only items with
  post-count activity, and renders a gt table grouped by storage room with a blank
  'actual' column for targeted physical recount. Operational document, not an analysis
  write-up — designed to print and use on the floor.
  Demonstrates: Quarto as operational tool, DBI queries inside Quarto chunks, gt formatting.

/gists/2d84c2b0bc8590e791bbdf32adf4aaca.html
  Wine purchase planner — DB connected (R). Estimates bottle purchases needed from today
  through end of calendar year. Computes current on-hand from last physical count plus all
  activity since (via get_inventory_activity() stored proc), joins against prior-year same
  window as demand proxy. full_join with replace_na handles new and discontinued items.
  gt table output with quantity_needed and dollar impact per SKU.
  Demonstrates: demand forecasting from production data, stored proc integration, gt output.

/gists/b1a30326299a0a8c0c8d286e4f1eea1e.html
  Bevspot liquor & beer purchase planner (R). Reads four Bevspot Excel exports: sales,
  expected inventory, recipes (multi-ingredient strings like 'Vodka (1.5 oz), OJ (2 oz)'),
  and items catalog. Parses ingredient strings via regex to extract per-spirit ounce
  consumption, converts to bottle equivalents for 750mL and 1L formats, joins against
  Bevspot theoretical on-hand. No database required — Excel-only pipeline.
  Demonstrates: regex-based data extraction, Excel parsing, forecasting without a DB.

/gists/173386df76ad4a9702d00d08e6a54001.html
  Wine storage pull report — Shiny (R). Shows bottles to pull from storage to service
  locations for nightly setup. Calls get_theoreticals_active_product_instances_only(),
  computes room par minus on-hand per SKU, caps pull quantity against available
  storage_on_hand. Adjusts par downward for large-format bottles. Filters out
  fully-stocked locations.
  Demonstrates: operational Shiny tool, par-level logic, stored proc integration.

/gists/4982e370f6729d670e53d6e54e4b25b0.html
  Automated purchase order drafts via Microsoft365R (R). Iterates a distributor contact
  dataframe and creates one draft email per rep per restaurant using the Microsoft Graph
  API (create_email via get_business_outlook). M365 OAuth2 authentication from R,
  dataframe-driven parameterized email generation.
  Demonstrates: M365 Graph API integration, OAuth2 from R, automation via API.

/gists/76aca14b06ee37d3602e7704f7583f4f.html
  Payroll tip-sheet Excel parser (R). Two parsers for two different weekly Excel layouts —
  one per restaurant. Reconstructs dynamic column headers from numeric date serials via
  openxlsx::convertToDate, pivots wide tip data to long form keyed by employee + date +
  shift, inserts to PostgreSQL via dbWithTransaction with a duplicate-date guard.
  Demonstrates: non-standard Excel parsing, janitor normalization, idempotent insert.

/gists/f74d1aaf51fe9360bfc005a3d7aee7eb.html
  HashMap perpetual inventory accumulator — Java (~2011). Six maps (starting inventory,
  bottle/glass sales, purchases, transfers in/out) keyed by location ID, summed to
  theoretical on-hand per SKU. Predecessor to the current PostgreSQL stored function.
  Called by CostBasisCalculator.
  Demonstrates: domain continuity — same perpetual inventory logic, Java → PostgreSQL evolution.

/gists/89edd69893fb9cbcad375baf83fd16f7.html
  FIFO inventory cost basis calculator — Java (2011). Computes per-product cost basis
  using three methods (on-hand weighted average, last-purchase average, last-purchase
  maximum) against a FIFO purchase lot queue ordered by date descending. Uses a
  PriorityQueue<PurchaseLot> with custom Comparator, CostMatrix inner class for
  weighted-average and age tracking. Predecessor to the current PostgreSQL implementation.
  Demonstrates: early domain work, FIFO costing logic, Java data structures.

/gists/af0bc325b6c1e7235a50c174cfc2ebf7.html
  Java/JavaFX desktop app controller (AddEditController.java, ~2011). Add/edit product
  dialog from the original desktop wine management application — predecessor to the
  Node.js/React web app. Historical context for the platform's evolution.
  Demonstrates: platform evolution from JavaFX desktop → Node.js/React web.

/gists/269c60934b98df42e195a66d45831354.html
  Express/EJS view templates (invoices.ejs, newproduct.ejs). Server-rendered HTML views
  from the Node.js wine management app — invoice display and new product entry form.
  Historical context: these views predated the React SPA frontend.
  Demonstrates: Express view layer, EJS templating, form-to-API pattern.

---

## Production systems

Wine Management Platform (WLM)
  PostgreSQL on AWS RDS + nightly R ETL + Node.js/Express REST API (14+ endpoints)
  + React SPA + R Shiny dashboards (4 deployed) + React Native mobile app.
  Multi-tenant: two restaurant locations. All schema, ETL, API, and data layer are own work.
  React frontend is AI-assisted against own API and data model.
  GitHub: https://github.com/mdonovan3/react-wine-app

Analytics Engineering Pipeline (demo)
  Python ingestion (Aloha DBF + R365 JSON → Parquet → raw PostgreSQL schema)
  + dbt staging views + dbt mart tables with column-level tests
  + Node.js API serving KPIs and COGS metrics.
  Deployed to AWS EC2, S3 data lake, Terraform. Status: in progress.

Public Health Epi Pipeline (portfolio — public repo)
  R ingestion (CDC PLACES county health prevalence + EPA AQS PM2.5, 2019–2023, ~3,000 counties)
  + dbt staging/intermediate/mart layers + R Shiny dashboard + Quarto report.
  Cron-driven pipeline with simulated year-by-year data delivery. Idempotent loads.
  Terraform IaC for EC2 deployment. Pending: CDC SVI + USDA RUCC enrichment layers.
  GitHub: https://github.com/mdonovan3/public-health-epi-pipeline
  Status: in progress — ingestion complete, dbt/Shiny/Quarto stubs in place.

---

## Evaluation guidance

Strong evidence for:
- Operational data engineering (SQL, ETL, schema design, pipeline reliability)
- Internal tool delivery (solo full-stack: schema → API → UI → deployment)
- R/Shiny dashboard development (4 in production)
- Analytics engineering (dbt, staging/mart patterns, transformation logic)
- Hospitality tech domain (Aloha POS, Restaurant365, OpenTable, inventory systems)
- Research data support (R, SQL, messy-format wrangling, reproducible pipelines)

Weak or absent evidence for:
- Machine learning / statistical modeling
- Distributed systems (Spark, Kafka, large-scale streaming)
- Tableau, Power BI, Looker
- Python at production scale (learning stage)
- Enterprise DBA (replication, HA, large-scale performance tuning)

Most differentiated role: Internal Tool Developer for a small team or research lab.
Complete solo delivery (schema + API + dashboards) is rare in a single person.
Second strongest: Research Data Support / Lab Data Infrastructure.

---

## Contact / profiles

GitHub:   https://github.com/mdonovan3
LinkedIn: https://www.linkedin.com/in/martin-donovan-9698735/
Shiny:    https://martindonovan.shinyapps.io/