Call Center Dashboard
Dynamic performance reporting — slicers, named ranges, and structured models that update as data flows in, tuned with XLOOKUP / LET / FILTER for speed at scale.
Download (Excel) ↓Systems dossier
Everything the homepage argues, itemized: seven system families, the public dashboards, the full toolkit, and the independent engineering. Optional depth, for the technically curious.
The build log
An exhaustive inventory of the systems behind the day job, catalogued for the technically curious — the note below explains the one thing it leaves out.
A connected suite of Manifest V3 extensions and single-file tools, engineered for a managed enterprise macOS Chrome environment and operating entirely from file:// origins.
Ingests a managed-Chrome policy export, resolves the developer-mode/extension-settings interaction, and computes the exact capability envelope available to unpacked extensions by subtracting blocked permissions and runtime-blocked hosts from the full MV3 permission set. The environment was mapped before a single tool was built.
An extension IDE, deployment pipeline, settings store, and bookmark manager in one self-contained HTML file run from file://. Manages multiple unpacked extensions, scaffolds new ones, and ingests whole-folder bundles through a fenced-block format with a name-match guard against accidental cross-paste. Writes real files via the File System Access API; autosave rides IndexedDB with signature comparison.
Grants the control plane Chrome-level reach from a file:// origin over a request/response envelope protocol: seven commands covering namespaced storage, full bookmark-tree read/write with favicon capture, and window/tab launching. A pinned manifest key fixes the extension ID; permissions are scoped to exactly three APIs.
A dual-FAB overlay injected across a single-page-app estate: route patching and MutationObserver unified into one onRoute(), a display wake-lock kept alive by a near-silent offscreen audio loop, scroll-capture-to-text, a local-tool launcher, copy/paste slots, a keystroke feed, and a global styles sheet propagated live across iframes and shadow roots.
An element picker with a persistent documentation table that records target identity per click — ignoring any injected UI — so every internal domain gets structured selector documentation that downstream extraction tools import.
Persists authenticated BI sessions across browser restarts by vaulting cookies across the analytics host and the SSO identity host and re-injecting on startup. A lost-update race from parallel storage writes was diagnosed and resolved by merging all cookies into a single write.
Crosses JavaScript world boundaries to read SAPUI5 binding contexts directly (a content script’s isolated world can’t see window.sap), fans extraction triggers across nested iframes, and returns payloads through a shadow-DOM sidebar. Four extraction modes: query-panel scrape, object/variable definition reads, coordinate-grid tables, and indent-stack tree reconstruction.
The house pattern underneath the whole fleet: self-contained offline HTML applications — no server, no dependencies, no data leaving the machine.
Each tool is one HTML file with embedded JSON data and in-browser XLSX parsing, built to run in a locked-down, file-origin environment. Every shipped file is simultaneously a snapshot and a re-loadable application.
A “save copy” mechanism that re-serializes the entire application with fresh data baked in — using a pristine-snapshot pattern and a dedicated script-tag injection point to dodge the classic bug where a self-editing download function’s regex matches its own source.
Aggregation and merge rules validated against system-of-record anchors, never assumed: rates roll up as sum-of-numerator over sum-of-denominator; in-progress periods are excluded from trends; incremental drops merge at day grain; columns parse by header name, never index; truncated worksheet ranges are recomputed before any parse.
CSS-token theming with zero hardcoded colors across the fleet — every tool ships integrated dark and light modes on the same data source.
Versioned Python builders assemble each tool: string replacement with exact match-count assertions, syntax checks on the extracted script, jsdom verification against pandas-computed truth values, and only baked outputs ship.
The flagship name withheld: one application, twelve views — the first thing ~40 supervisors and managers open each morning.
Performance, efficiency, attendance and adherence with a full auxiliary-state breakdown, coaching, alerts, calendars, per-advisor scorecards, daily ledgers, and an embedded outlier-triage view across hundreds of agents — all in one offline file with search, frozen headers, threshold filters, and XLSX/JSON/CSV export.
The builder ingests raw scheduling, HR, device, and roster files and synthesizes the complete dataset entirely in the browser — retiring the manually assembled workbook it replaced.
A full pay period rendered as parallel per-day activity columns at minute resolution, anchored by clock-in/out bands, with a restrained palette that reserves a single accent for attention states — so problems surface before you consciously look for them.
Structured coaching forms that preserve the six-week metric window exactly as it looked at each past session, filter the coach’s own prior wording in real time as they type, and close every record with a before/after metric delta. Serial-numbered and audit-ready.
Per-advisor hover analytics including a leave-one-out team-impact figure: the points an advisor moves the team average versus the team without them — one hover away, never in the way.
Login/logout time reconciled against payroll time per advisor per day, feeding a daily investigation drill distributed to supervisors — every flagged discrepancy is investigated and confirmed with the person before any correction is made.
Dozens of recurring analytical workbooks on a template-and-archive cadence, plus the forensic tooling that keeps them honest.
LET, LAMBDA, FILTER, BYROW, HSTACK, CHOOSECOLS, SEQUENCE and spill logic as the standard formula stack, with structured-table XLOOKUP joins over a shared roster backbone common to every engine.
A schedule exporter that reshapes a 31,000-row workforce-management activity dump into readable per-person weeks; an early-warning attendance-governance workbook with escalation templates; an on-time-plotting audit that generates WFM change requests; a supervisor-effectiveness stack-ranking model with a coaching-effectiveness index; billable-training calculation pipelines; compliance controls for security attestations and refund reviews; quarterly absenteeism and attrition analyses.
A structural x-ray of any workbook: eighteen toggleable audit categories — formulas, names, pivots, conditional formatting, external links, a full VBA inventory — each wrapped in its own error trap so one failure never aborts a run. Chases external links and audits linked workbooks read-only; renders a themed report sheet plus an exhaustive JSON export.
Exports every VBA component with a JSON manifest so the extension-side editor can edit Excel code externally, then applies staged changes back — in-place replacement for document modules, remove-and-reimport for the rest — with a written apply log.
A documented set of Mac Excel bug mitigations: comment-shape realignment, note autosizing by wrapped-line estimation, threaded-comment migration that preserves authorship — and a deliberate one-line macro that clears the undo stack to reclaim memory on a taxed machine.
Two-million-row datasets held connection-only on a platform with no Data Model support: buffered single-pass cardinality diagnostics, distinct-first classification against millions of rows, and dynamic column derivation so schema drift never silently drops a column.
The pipeline and query engineering the whole tool fleet stands on.
A canonical query folder as the single upstream for every downstream tool: interaction, quality-evaluation, survey, and peer-feedback pulls alongside adherence, punch, roster, and scheduling extracts, refreshed on a weekly production cadence.
Merged reports joining six data universes into one row per interaction across 235 documented variables — built on a gatekeeper-variable pattern with strict context operators, plus a discovered list of platform constraints that shaped every formula.
Pairwise correlation matrices computed in-formula (LET + MAP over SEQUENCE, with SUMPRODUCT pair counts), chi-squared evidence analysis, and audit-grade sample sizing — n₀ = p(1−p)(z/E)² with finite-population correction and two-sample significance testing.
Retail-style 5-4-4 fiscal period math threaded through every week code in the reporting estate, cross-checked against the calendar’s own arithmetic across test anchors.
Scheduled report emails captured by cloud flows, transformed locally by scheduled Python builders over synced document libraries, then published and distributed with download-forced share links — an end-to-end reporting workflow with no manual step.
The measurement and evidence machinery behind the site’s May 2026 COPC recertification — built, not bought.
A single-file application that became the program’s system of record: a searchable standards codex, a meetings log with heat-shaded calendar rails and verbatim transcript drawers (name-scrubbed), corrective-action tracking against real item numbering, and built-in calculators — behind a macOS-style global search overlay.
A mandated evidence structure per corrective action — observation, actions, evidence, compliance case — under strict authorship rules: other owners’ wording preserved verbatim, nothing fabricated, gaps flagged rather than filled.
Replaced a flattering but non-representative measure with QA-observed audits: formal sample sizing, a three-module framework (sizer, yield, actuals tracker), a collaboration form feeding the tracker, and a data spine at case-and-advisor grain whose defect flag is never null.
Exactly two sanctioned derivation methods; a gap-based customer-satisfaction method with scheduled refresh; a calibration scatter that exposed a structural impossibility between two paired targets (complementary tails of one distribution); utilization redefined to in-office shrinkage; trailing-twelve-month baselines with a delta projection bridging missing history.
Schedule attainment rebuilt from an aggregate ratio that could exceed 100% into a per-interval pass rate; a twenty-item discrepancy checklist with exact artifact, page, and cell locations; facilitation guides with run-of-show timing, roster chips, and domain-split question banks.
A multi-phase corrective program on the site’s top dissatisfaction driver: a multi-report join measured in tens of megabytes, a hypothesis-testing register with proven/disproven outcomes, weekly monitoring books, and behavior-level correlation analyses joining quality evaluations to surveys.
The glue layer: native launchers, document engineering, and disk archaeology.
Desktop app bundles that wrap the file-origin tools as chromeless native launchers — resolving a windowing constraint no extension could — including a dual-tenant messaging split and process-kill utilities.
A 3,102-entry (~11 GB) filesystem manifest rendered as a dark file-tree browser with proportional size bars, lazy expansion, five-state flagging with per-node comments, and JSON export — used to reconstruct an entire analytical workstream from disk archaeology.
Fragmented chat screenshots reconstructed into clean verbatim transcripts, then compiled into interactive PDFs: structured minutes with numbered footnotes and bidirectional anchor navigation between every claim and its source line.
A master dashboard housing multiple analytics tools as registered iframe components — a four-edit registration pattern, persistent mounts, and a message-protocol bridge feeding live data to each tool.
Public work
Built independently from public datasets, with the same principles I apply in a professional operations environment: clarity, utility, and actionable insight. Each ships with integrated light and dark modes on a single data source.
Dynamic performance reporting — slicers, named ranges, and structured models that update as data flows in, tuned with XLOOKUP / LET / FILTER for speed at scale.
Download (Excel) ↓Interactive data storytelling over a rich public dataset — compelling charts, conditional formatting, and drill-downs that make patterns legible at a glance.
Download (Excel) ↓Geographic and time-series analysis with KPIs, sparklines, and dynamic text — a clean layout and reusable components that streamline both build and daily use.
Download (Excel) ↓Toolkit
Dynamic-array Excel (LET, LAMBDA, FILTER, XLOOKUP) · Power Query M · VBA + workbook forensics · SAP BusinessObjects WebI · ETL (BusinessObjects, Superset, Power Automate, SharePoint) · applied statistics
Single-file offline web apps (HTML / CSS / JavaScript) · Chrome extensions (Manifest V3) · Python build pipelines · UX-oriented tool design
COPC CX Standard (R7.1 / R8.0) · DMAIC · target-setting governance · quality monitoring & coaching frameworks · WFM (NICE / IEX), adherence, forecasting
People management & weekly coaching cadence · supervisor onboarding & development · new line-of-business launch · stakeholder & client communication
Fluent on Windows and macOS · Office 365, Adobe Creative Cloud. Technically minded with an eye for design and frictionless UX — off-hours in C++, Python, and Assembly.
Off the clock
Self-directed builds I chase past the point of reason — some for a weekend, some for many years.
Drive an iPad from a Windows PC over one Bluetooth radio — keyboard, mouse, audio, and clipboard — through a headless Debian VM that owns the radio and pretends to be a BLE keyboard. A deep dive into Bluetooth Low Energy, USB passthrough, and PipeWire audio, packaged to a single Windows exe.
Read the field manual → Source on GitHub →A Chrome extension that reshapes the browser around how I work. One blue button on every page carries a universal toolkit; a red button reacts to whichever site you’re on — and a built-in Site Autopsy tool documents any new site into a spec that generates its own companion. Distributed as a single-file build console with a live feature matrix.
Read the field manual →Reverse-engineer a 25-year-old, closed-source game engine by hand, in x86 assembly — no source — and rework it from the inside out. Seven years of RE consolidated into one version-guarded runtime DLL: 86 assembly patches, 36 custom maps, 60,000+ recolored sprite frames, and 2,100+ data-table edits.
Read the build notes → Contact for more info →Tile every surface you live in side by side — each pane a real Firefox window, not a webview — reparented into one frame. Import your Firefox profile once and every pane is signed in; a single omnibox drives whichever pane you’re in, and Firefox’s own chrome hides until you want it. A leash of foxes is three; this is a leash of Firefoxes. Packaged to a single Windows installer.
Read the field manual → fireleash.com → Source on GitHub →Let’s talk
References available upon request.