Santiago Saca

Business Data Consultant · Builder

Miami, Florida

I fix enterprise data by day — and build the tools I wish existed by night.

Nearly four years in Genpact's Business Data Services practice taught me how product, vendor, and customer data actually breaks inside Fortune 500 systems. So I started shipping the software I kept wishing my clients had: an AI-assisted master-data platform, a cleansing engine for SAP migrations, a live real-estate marketplace for Guatemala — and one very ambitious fishing game.

Santiago Saca smiling outdoors at the Grand Palace in Bangkok

Santiago Saca Miami, FL · English & Spanish · M.S. Information Systems, Indiana University Currently building: Lustro & The Veil Strand

View projects Get in touch
GitHub LinkedIn
5Products designed & built solo
4Live in production today
90k+Lines of code across projects
4Industries served at Genpact

Selected work

Five products, built end to end.

Everything below was designed, engineered, and built by one person. All five share a thesis: the discipline I sell as a consultant — clean data, governed workflows, audit trails — is worth building directly into software.

No. 01Enterprise master data

iDOT PIM

An AI-assisted master data platform with enterprise-grade approval workflows, compliance routing, and segregation of duties.

iDOT manages the three core enterprise master-data domains — vendor, customer, and product — from first request to approved record. Every change runs through a configurable approval workflow with SLA tracking, audit timelines, and compliance branches for LATAM, EU, SOX, and GxP rules. It is the exact discipline I design for Fortune 500 clients, implemented as working software.

  • Context-aware workflow engine inserts approval steps at runtime: corporate security review for LATAM vendors, finance sign-off for the EU, supervisor review whenever SOX-sensitive fields like bank accounts are touched.
  • Segregation of duties enforced in code — no one can approve their own request, including the master-data team itself.
  • Claude-powered document AI: upload a W-9 or VAT certificate and the onboarding form fills itself, with an "AI-populated" badge on every field for the audit trail.
  • Weighted fuzzy matching across name, tax ID, DUNS, and address stops duplicate vendors before they are ever created.
  • Roughly 21,700 lines of code — 109 routes, 27 database tables, 13 demo roles — built in ten days.
Node.jsExpressSQLiteClaude APITailwind CSSDockerFly.io
Live site
iDOT PIM's no-code workflow configurator showing trigger rules for a LATAM escalation scenario
The no-code workflow configurator. Admins build trigger rules against form fields that are auto-discovered from the live form.
iDOT vendor onboarding request with an AI document-upload step that reads W-9s and pre-fills the form
AI document upload: Claude reads W-9s and pre-fills the form.
iDOT role-based dashboard with KPIs and quick actions
Role-based dashboards with KPIs and quick actions.
No. 02Data cleansing for ERP

Lustro

AI-powered master-data cleansing that turns dirty vendor lists into SAP-loadable files — with a human in the loop only when confidence is low.

ERP migrations still run on consultants hand-cleansing spreadsheets for weeks. Lustro takes messy vendor, customer, product, or finance master data, cleanses and standardizes it, and outputs the exact load format the target ERP requires — SAP S/4HANA first, with Oracle and Dynamics 365 profiles. A deterministic rules engine handles the predictable majority; AI touches only the ambiguous residue under a hard $2-per-job budget cap.

  • Generates the real SAP Migration Cockpit package — the per-structure CSV-ZIP format most teams assume is a simple Excel file. It isn't, and Lustro builds it correctly.
  • Sanctions screening built in: vendors and customers checked against 71,095 bundled OFAC and EU entries with two-tier exact and fuzzy matching.
  • Tamper-evident by design: an append-only, hash-chained audit log enforced by a database trigger, with strict per-customer data isolation.
  • A hard safety guardrail: fuzzy matches never auto-merge, so a confidently wrong merge of two legal entities can never reach the ERP.
  • Live with self-serve signup and a free tier — with Stripe billing and two-factor login built in, about 15,500 lines across a rules engine, API, batch worker, and React app.
PythonFastAPIPostgreSQLClaude APIReactTypeScriptStripeFly.io
Live site
Lustro landing page with the headline 'Clean your master data for any ERP, with AI'
The self-serve front door. Upload a file, map the headers, and export an ERP-ready package.

Try it — edit a name. A miniature of Lustro's matcher: legal suffixes stripped, names scored, and even a 100% fuzzy match only ever routes to review — never auto-merges.
No. 03Consumer marketplace · Live at viveguate.gt

ViveGuate

Guatemala's Zillow: the country's property listings on one map — deduplicated, and priced with real comps.

ViveGuate's split-screen search: listing cards on the left, a Google Map of Guatemala City with clustered price pins on the right
The split-screen search. One canonical record per home, no matter how many sites list it.

Guatemala had no Zillow — inventory was scattered across dozens of broker sites and classifieds, with no unified search and no price transparency. ViveGuate aggregates listings from about 40 sources into one deduplicated map, layers on an automated valuation model with transparent comps, and wraps it in a full marketplace: accounts, offers, messaging, and email alerts.

TypeScriptNext.js 15PrismaPostgreSQLGoogle MapsClaude APIVercel
Live site
Try it — drag a point. How zone pricing gets its shape: the same Voronoi tessellation ViveGuate clips to municipal boundaries, computed live in this figure.
  • 8,000+ live listings from 39 source adapters, kept clean by 30 production cron jobs running staged ingestion waves and nightly data-quality passes.
  • A cross-source deduplication engine — blocking, deterministic scoring, AI tie-breaks only on the gray zone — built on the principle that a false merge is worse than a missed duplicate.
  • A valuation model that shows its work: anonymized comps behind every estimate, uncertainty bands that widen honestly, and backtests that grade the model's accuracy.
  • Computational geometry in production: zone price polygons generated with Voronoi tessellation, clipped to official municipal boundaries.
  • About 44,700 lines of TypeScript shipped solo to a real domain — legal pages, one-time-code login, SEO, and a 131-finding audit closed out through a four-gate checklist.
No. 04Game development

The Veil Strand

A solo-built fishing life-sim where every story beat is something you catch — run with studio-grade data discipline.

My dream game, in active development for Steam. You inherit a tackle shop in Carillon, a harbor town that gives its unbearable memories to the sea; on rare "Silvered tides" the water turns mirror-still and you can fish the past back, one fragment at a time. Fishing is simultaneously the economy and the story.

It is also a deliberately engineered production. The first prototype was retired with a written eight-cause postmortem, and the reboot runs on CI-enforced quality gates, a fully data-driven architecture, and complete asset provenance.

  • A deterministic nine-state fishing engine with seeded randomness, so any session can be replayed exactly — backed by 23 dedicated unit tests inside a CI deploy gate that runs 89, with a standing law that no failing build ever merges.
  • Every balance number lives in data files, never code, and all 324 generated or purchased art assets carry a machine-readable provenance record audited by CI.
  • An AI art pipeline locked to an 886-color master palette with a per-pixel audit, so generated art can never drift off-style.
  • 151,000 words of design documentation — story bible, art direction, and an 18-month roadmap with budgets and risk registers.
  • Playable today: a free-walk town build with a day-night cycle, animated water, and the signature full-screen Silvered-tide shader.
Pixel-art screenshot of the harbor town Carillon at noon
Noon. Carillon on an ordinary day.
Pixel-art screenshot of Carillon at dusk, windows glowing with warm lamplight
Dusk. Lamplight on the harbor.
Pixel-art screenshot of Carillon during a Silvered tide, the water mirror-still and silver
The Silvered tide. The game's signature moment — the sea goes mirror-still.
Godot 4.6GDScriptPython toolingGitHub Actions CIPixelLab APIElevenLabs APICustom shaders
In development — Godot 4
No. 05Client work

Lumetric Films

A cinematic portfolio and booking site, designed and built single-handedly for a real video production studio in New York.

A working videographer had real clients but no professional web presence. I acted as consultant and developer: gathered requirements, structured the studio's videos and pricing into one clean data model, and shipped a polished site the owner updates by editing plain data files — no developer needed for routine changes. First commit to live deployment in about two weeks.

  • A custom live-preview engine: each portfolio tile silently loops a hand-picked segment of its video, spun up only when the tile scrolls into view and torn down when it leaves.
  • A theater-mode lightbox with category playlists, keyboard navigation, and a properly focus-trapped dialog.
  • Everything is data-driven: 12 portfolio videos, 5 categories, and 19 priced offerings live in plain data files the client can edit themselves.
  • Accessibility taken seriously for a marketing site — focus traps, screen-reader labels, and full reduced-motion support.
  • Real stakes: live pricing, a working inquiry pipeline to the owner's inbox, and verified production work for named clients.
React 19ViteTailwind CSS 4YouTube IFrame APIWeb3FormsVercel
Live site
Lumetric Films portfolio section showing live performance work on a cinematic dark layout
The portfolio grid. Every tile is a live, muted preview — no static thumbnails.
Lumetric Films homepage hero, dark and cinematic with bokeh and the studio wordmark
The hero, with a sequenced entrance animation built around a hand-drawn SVG wordmark.

Method

Rules I build by.

Five independent systems kept teaching the same lessons. At some point they stopped being habits and became laws — each one enforced by a named mechanism, not a slogan.

Experience

Where the domain expertise comes from.

The day job is the research lab. Nearly four years inside global master-data programs is why the products above understand compliance routing, survivorship rules, and segregation of duties.

Business Data Consultant

Genpact — Business Data Services

Aug 2022 – Present · Miami, FL
Life Sciences & Healthcare · Food & Beverage · CPG · Manufacturing

  • Authored the functional Business Requirements Document for a global Supplier Information Management system — smart workflows, embedded regional compliance, ERP integration, and third-party risk assessment.
  • Wrote user stories and functional specs for a global Product Information Management system, including automated component reuse and multi-variant product creation in a single workflow.
  • Led data quality workstreams end to end: maturity assessments; profiling across completeness, uniqueness, validity, accuracy, and consistency; and remediation across customer, vendor, and product domains.
  • Defined data quality KPIs and built a Power BI dashboard tracking product master data quality across LATAM.
  • Partnered with business and technical stakeholders to design target-state operating models, end-to-end process maps, and fit–gap analyses against industry best practices.

Skills

The toolkit, in three layers.

Master Data & Delivery

  • MDM — Customer / Vendor / Product
  • PIM & Supplier Information Management
  • Data Governance & Data Quality
  • Data Maturity Assessment
  • Fit–Gap Analysis & E2E Process Mapping
  • BRDs, User Stories, Stakeholder Workshops

Data & Analytics

  • Data Profiling
  • Data Cleansing / Deduplication / Enrichment
  • Power BI
  • KPI Design

Tools & Tech

  • Python & SQL
  • Snowflake & PowerMe (data profiling)
  • Git
  • Next.js / TypeScript — shipped full-stack products
  • Node.js
  • Godot / GDScript

Education

Indiana University Bloomington.

Both degrees are from IU — and I'm preparing to go back for the next one. The projects above taught me data architecture, entity resolution, and AI reliability in practice; graduate study is how I put deeper theory underneath them.

2022

M.S. Information Systems

Indiana University Bloomington

2021

B.S. Business Administration

Indiana University Bloomington

Contact

If any of this is useful to you, say hello.

I'm happy to talk about master data, the products above, graduate study, or what it takes to ship software solo. English or Spanish — whichever you prefer.

santiagosaca23@gmail.com
LinkedIn GitHub Miami, FL English & Spanish