elwin@sytadel:~ $ ▋

Elwin Ernst

Elwin Ernst

Software Engineer · Security & Identity · Applied AI

4+ years building and operating real production systems — multi-tenancy, authentication, LLM agents. Based in Montevideo, Uruguay.

  • available for work
  • 4+ years
  • Montevideo, UY

elwin@sytadel:~$ # 4 records ▋

  1. Asociación Argentina Central

    Product / Software Engineer

    Design and development of administrative systems and multi-module internal platforms. Built sales and inventory systems and process-automation tools. Web and mobile apps with direct end-user work, plus data integration and reporting for the organization's decision-making.

  2. Xirect Software Solutions

    Full-Stack Engineer

    Built a logistics and shipping web platform (ShippingMadeEZ) with real-time tracking, user and role management, package labeling and cost calculation across carriers. Also a web and mobile platform for Fuxion distributors: training, sales, carts, events, calendars and video calls.

    view platform
  3. Unión Uruguaya

    Software / Data Engineer

    Internal tooling: a web budget system (migrating an 80-sheet Excel into reporting with drill-down), a Business Intelligence platform (Databricks + Power BI versioned as code), an internal commissions system, colporteur tracking (SEHS) and an intranet.

  4. InmoSimple

    Co-founder · Full-Stack Engineer

    End-to-end full-stack real-estate SaaS: landing, admin panel, backend, frontend, property scraper and deployment infrastructure.

    view platform

elwin@sytadel:~$ # services ▋

proc — projects 7 tasks · 4 public
public 4/7
with demo 2/7
on npm 1/7
  • Security and identity platform for teams building with AI agents.

    • TypeScript
    • NestJS
    • PostgreSQL
    • Docker
    • Modular control plane: auth, billing, vault and zero-trust as independently deployable services.
    • Agent identity, Model Context Protocol and human approval of sensitive actions.
    • Several of its services already run in production (such as auth-api); the full suite is still in development.
  • Multi-tenant identity provider: the foundation of Sytadel's trust plane.

    • NestJS
    • TypeORM
    • PostgreSQL
    • Refresh-token rotation with reuse detection.
    • Passkeys / WebAuthn resistant to account enumeration, tenant-scoped JWT (OWNER/ADMIN/MEMBER).
  • sytadel-mcp-server in progress public

    Model Context Protocol server that exposes Sytadel's identity and access primitives.

    • TypeScript
    • MCP
    • Node.js
    • Drivable from Claude Desktop, Cursor and any MCP-compatible client.
    • Exposes identity and access control as tools an AI agent can invoke.
  • @elwinernst/nest-auth library public

    Published npm library: reusable JWT verification for the identity plane.

    • NestJS
    • TypeScript
    • npm
    • Replaces duplicated JWT-verification implementations across several Sytadel services.
    • Published publicly on npm (v0.3.0).
  • mango case › personal use private

    Agentic personal finance for LATAM: AI is the interface, not forms.

    • Go
    • Claude
    • PostgreSQL
    • You talk in natural language with an assistant (Claude) that reads and writes your financial data via tool-calling.
    • No forms or dashboards: LATAM-focused design with subscription handling.
  • nuvia in progress private

    Multi-tenant management SaaS (scheduling, inventory, payroll).

    • NestJS
    • Next.js
    • PostgreSQL
    • TypeORM
    • Multi-service appointment scheduling with clients and staff, inventory and a public catalog.
    • Mixed payroll: commission, base salary and hourly pay.
  • inmosimple product private

    End-to-end full-stack real-estate SaaS, co-founded and in production.

    • Next.js
    • NestJS
    • PostgreSQL
    • Landing, admin panel, backend and frontend, end to end.
    • Property scraper and deployment infrastructure.

elwin@sytadel:~$ # what I offer ▋

Four lines of work, all backed by systems I built and operated in production. No inflated promises: concrete problems, concrete solutions.

$ services --dev

Software development

You need a product or an internal tool, but off-the-shelf software doesn't fit your operation and doing it all by hand doesn't scale.

What I offer

  • ▸ Full-stack web apps and multi-tenant SaaS platforms.
  • ▸ Custom management systems (sales, inventory, scheduling, payroll).
  • ▸ APIs and integrations between services and internal tools.
  • ▸ Desktop and mobile apps when the case calls for it.

Examples

  • ✓ End-to-end real-estate SaaS: landing, admin panel, backend, scraper, and deployment infrastructure.
  • ✓ Logistics platform with real-time shipment tracking, roles, and cost calculation across carriers.

For whom
Companies that need their own product or want to digitize an operation that today lives in spreadsheets and manual processes.

$ services --arch

Architecture & consulting

The system works, but security, multi-tenancy, or scalability start to hurt — and a bad architecture call is expensive to undo.

What I offer

  • ▸ Backend architecture (clean / hexagonal) built to last.
  • ▸ Authentication and authorization: JWT, RBAC, passkeys, refresh-token rotation.
  • ▸ Multi-tenancy and data isolation with Row-Level Security.
  • ▸ Database design, application security, and scalable integrations.

Examples

  • ✓ Multi-tenant identity provider with refresh-token rotation and reuse detection, enumeration-resistant passkeys, and tenant-scoped JWTs.
  • ✓ Modular security control plane with auth, billing, and vault as independently deployable services.

For whom
Technical teams starting a new product or needing a second opinion on identity, security, and scalability.

$ services --ai

Applied artificial intelligence

You want to leverage language models in your product or operation — but not a toy chatbot: genuinely integrated, secure, and under control.

What I offer

  • ▸ Language-model integrations inside your own systems.
  • ▸ Custom assistants with scoped access to your data.
  • ▸ Agents with tool-calling and the Model Context Protocol (MCP).
  • ▸ Intelligent automation and AI-driven information processing.

Examples

  • ✓ Agentic finance assistant where the AI reads and writes the user's data via tool-calling — the AI is the interface, not the forms.
  • ✓ MCP server exposing identity and access primitives as tools an agent can invoke from Claude Desktop or Cursor.

For whom
Companies and product teams that want to adopt AI practically, with clear boundaries and human validation.

$ services --automation

Business automation

Your team loses hours to repetitive admin tasks, manual reports, and data scattered across spreadsheets.

What I offer

  • ▸ Automation of admin flows and repetitive operational tasks.
  • ▸ Reporting and dashboards for decision-making.
  • ▸ Data integration across systems that don't talk to each other today.
  • ▸ Internal tools that cut manual work.

Examples

  • ✓ Migration of an institutional budget from an 80-sheet Excel file to reporting with drill-down.
  • ✓ Business Intelligence platform versioned as code, plus an internal commissions system.

For whom
Organizations with heavy manual processes that want to free up their team's time and decide with reliable data.

Let's talk about your case →

Tell me what you need and we'll see if I can help. No commitment.

elwin@sytadel:~$ # AI for teams ▋

I help companies and teams adopt AI tools, automate repetitive tasks, and improve productivity through practical, secure applications. AI applied to real work, not to the demo.

Formats

$ training --intro

Introductory training

Generative-AI fundamentals, tooling, prompting, information analysis, and best practices.

For whom: Teams just starting out that want a solid, safe foundation.

$ training --teams

Team-based training

Workshops tailored to admin, finance, HR, sales, customer support, and operations.

For whom: Areas with concrete tasks where AI can cut manual work.

$ training --audit

Assessment & automation

Identifying repetitive tasks, analyzing opportunities, designing flows, and building prototypes.

For whom: Organizations that want to move from theory to flows that work.

Method

  1. 01

    Assessment

    We understand the team's real work and where AI adds concrete value.

  2. 02

    Training

    Hands-on training tailored to each area's tasks and tools.

  3. 03

    Practical application

    Designing flows and prototypes on the team's real cases.

  4. 04

    Follow-up & improvement

    Support, usage measurement, and tuning of what was implemented.

Privacy & responsible use

  • ✓ Privacy and data protection
  • ✓ Human validation of results
  • ✓ Known model limits and errors
  • ✓ Company information security
  • ✓ Responsible use of AI tools
Request information →

Let's set up a consultation to see which format fits your team.

elwin@sytadel:~$ # stack ▋

security-and-identity/

  • JWT
  • RBAC
  • multi-tenancy
  • WebAuthn / passkeys
  • refresh rotation
  • Row-Level Security
  • AES-256-GCM
  • rate limiting

applied-ai/

  • Anthropic SDK
  • LangGraph
  • Model Context Protocol
  • tool-calling
  • agents
  • Claude Code

backend/

  • Node.js
  • NestJS
  • Express
  • Go · Chi
  • Deno

languages/

  • TypeScript
  • JavaScript
  • Go
  • SQL
  • Python
  • C#

frontend/

  • React
  • Next.js
  • Astro
  • Tailwind
  • Radix UI
  • TanStack

data/

  • PostgreSQL
  • MySQL
  • TypeORM
  • Prisma
  • Drizzle
  • sqlc
  • Supabase
  • Neon

devops-and-testing/

  • Docker
  • docker-compose
  • GitHub Actions
  • GitLab CI
  • Fly.io
  • Vercel
  • Jest
  • Cypress
  • testcontainers

elwin@sytadel:~$ # activity ▋

Public GitHub data, as of 2026-11-17

Contributions (12 mo)
1,949
Commits
1,264
Pull requests
484
Issues
174
LessMore

Recent activity

Featured repositories

  • sytadel-suiteTypeScript★ 1

    Meta-repo de Sentinel Suite: docker-compose, smoke tests, sentinel-app (Next) + sentinel-web (Astro) y submódulos auth-api, zerotrust-api, securechain-vault, billing-api.

  • auth-apiTypeScript

    Authentication service (JWT, sessions) for the Sentinel Suite

  • Model Context Protocol server exposing Sytadel identity + access primitives — driveable from Claude Desktop, Cursor, and any MCP-aware client.

  • zerotrust-apiTypeScript

    Zero-trust authorization service for the Sentinel Suite

  • Secure secrets & credential storage vault for the Sentinel Suite

View full profile on GitHub →

elwin@sytadel:~$ # open source ▋

Contributions to external open-source projects: real-problem research, technical root cause, and a tested solution. The value is in the engineering, not in the recognition.

status
Real status: all three are open proposals awaiting review by the TypeORM maintainers. None has been merged or reviewed yet.

Open proposals (awaiting review)

  • typeorm/typeorm#13000 +186 -42

    Problem
    OrmUtils.deepCompare runs on every save() for each json/jsonb and array column; on unchanged values (the common case when saving a loaded entity) it traversed the whole value ~8–14x slower than a straightforward deep-equal.
    Root cause
    It ran two full passes per object, each with two hasOwnProperty and two typeof checks per key, plus repeated lookups.
    Solution
    Rewritten to a single pass with a positional fast path for arrays, preserving exact semantics (Buffer, Uint8Array, Map/Set, Date/RegExp, NaN, and cycle detection).
    Verification
    200,000-case differential fuzz against the previous implementation: 0 divergences. Unit tests expanded from 3 to 12 cases.
    Benchmark
    20,000 objects: 63.8 ms → 7.7 ms (~8.3x); consistent ~6–8x at smaller sizes.
  • typeorm/typeorm#13001 +443 -3

    Problem
    insert().orIgnore() (and orUpdate paths that skip a row) wrote the RETURNING rows onto the wrong entities.
    Root cause
    Rows were matched by array position; with ON CONFLICT DO NOTHING a skipped row returns no RETURNING row, so everything after it shifts up by one entity (ids and generated values crossed).
    Solution
    Match each returned row to the entity that produced it by its identifying columns (primary + unique) instead of by position; with no skipped rows the previous mapping is kept.
    Verification
    Reproduced on PostgreSQL; skip-first/middle/last, multi-skip and no-skip cases. Regression tests running on postgres, cockroachdb and better-sqlite3.
  • typeorm/typeorm#13029 +314 -8

    Problem
    upsert() mutated the caller's input objects in place.
    Root cause
    On drivers without RETURNING (mysql/sqlite) the primary key was fabricated from the insert id and merged into each entity; on updated rows that id is wrong → silent corruption of the in-memory objects, with no error or warning.
    Solution
    When the operation is an upsert, operate on clones (preserving the entity prototype) and leave the originals untouched; generated values remain available on the InsertResult.
    Verification
    Tests that fail on master and pass with the change (3 cases); also verified against sqljs; the existing upsert/returning suites still pass.

elwin@sytadel:~$ ▋

███████╗███████╗
██╔════╝██╔════╝
█████╗  █████╗  
██╔══╝  ██╔══╝  
███████╗███████╗
╚══════╝╚══════╝

elwin@sytadel

───────────────────────

role
Software Engineer (graduated 2026)
host
Montevideo, Uruguay
uptime
4+ years in production
kernel
TypeScript · Node.js · NestJS
focus
Security & Identity · Applied AI
shell
zsh · tmux · vim
edu
B.Sc. Information Systems Engineering — UAP
langs
ES (native) · EN (conversational)
off
lifeguard · scuba diving
net
github.com/ElwinErnst

elwin@sytadel:~$ # let's talk ▋

Got a project, a technical question, or want to train your team? Drop me a line. I reply in Spanish or English.

Direct channels

$ contact --reason

The button opens your mail client with the message ready — nothing is stored on any server. Or just write to the email above directly.