Data diagnosis
A source-to-destination map, sample reconciliation and a classification of the defects.
DATA / INTEGRITY
Data pipelines & catalog quality
Make the data dependable before building on top of it.
I trace how records are created, matched, updated and reconciled. The work can include product normalization, fiscal enrichment, duplicate repair and rebuilding a sync so the same error does not return on its next run.
Discuss this service
Is this the problem
you are facing?
- Reports disagree with source records and nobody can explain the difference.
- Product identifiers, descriptions or country-specific fields are inconsistent.
- Retries, pagination or imports create duplicate records or partial updates.
A PRACTICAL ENGAGEMENT
What the work can include
Repair and validation
Explicit matching rules, a reversible repair approach and checks for missing or duplicated records.
Reliable processing
A repeat-safe import or enrichment path, exception reporting and reconciliation against a defined source of truth.
EVIDENCE / PUBLISHED WORK
See how I have approached it
The catalog study documents a 243,000-SKU enrichment API and 26,420 corrected rows across 16 defect categories. Its runtime uses exact, structured and normalized-description lookups without LLM calls. The payment study traces duplicate records to pagination and describes an idempotent replacement.
Data pipelines
243,000 SKUs, Zero LLMs at Runtime
A deterministic fiscal-enrichment API for a 243k-SKU auto-parts catalog: a 3-stage lookup chain, 26,420 rows corrected, zero LLM calls at runtime.
Node 20 · Express · PostgreSQL 16 · Redis (ioredis) · Zod
Data pipelines
The Numbers Were Wrong — AP Data-Integrity Rescue
Tracing an eight-figure AP overstatement to a duplicated pagination parameter: 2,725 surplus rows purged, the sync rebuilt as an idempotent pipeline.
React 18 + Vite · Keycloak OIDC · Node + Express · NocoDB · n8n
Data pipelines
Enterprise Inventory for a LATAM Tire Retailer
A 4-tier RBAC inventory system for 5,000+ SKUs across 4 warehouses: dual auth, immutable audit logging and mobile-first floor use at 375px.
React 18 · Vite 5 · TanStack Query v5 · keycloak-js 26 · Express 4
Systems architecture
Polling on the ETA, Not the Clock — Meridian
Meridian: a multi-tenant container-tracking SaaS on PostgreSQL RLS with an adaptive, ETA-anchored polling scheduler that cut vendor API calls ~6×.
Node 20 · Express + Helmet · PostgreSQL 16 (Force RLS) · Redis 7 · React 18 + Vite
BEFORE WE START
Common questions
Do you need the entire database to estimate the work?
An anonymized sample, schema and description of the failure are usually enough to begin discovery. Production access is scoped separately.
Can AI clean a product catalog?
It can assist with ambiguous text, but identifiers and fiscal mappings need validation and traceability. The published catalog runtime deliberately uses deterministic matching.
How do you know a repair worked?
Define reconciliation checks before the change: record counts, duplicate keys, known examples and relevant totals. Keep exceptions visible instead of silently discarding them.
START WITH A $150 AUDIT
Where is your business losing
leads or hours?
Tell me what happens today and what keeps going wrong. I reply with times for a 45-minute audit call; you get a written fix list with fixed prices, and the $150 is credited against any build.