Skip to content
View rdorta27's full-sized avatar

Block or report rdorta27

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
rdorta27/README.md

Rubén A. Dorta A.

Solution Architect | Applied AI & OSS/BSS

I build AI agents that respect business rules, the way I built telecom systems for 25 years.

I turned telecom operators' business rules into production systems for 25 years: product catalogs, ordering, charging and billing, integrated end to end. The last ten were at Ericsson, embedded with Claro Colombia. Today I build AI agents, RAG systems and evaluations in Python.

Projects

Flow Audit

An audit workbench for BSS order flows: it reads evidence exported from the catalog, OMS, BPM and provisioning, and reports where orders get stuck.

  • Today: dataset, architecture and rule catalogue in place; the audit engine is next.
  • Key decision: it observes evidence exports instead of touching the systems it audits, and reconstructs each order's journey into one canonical event stream before any rule fires.
  • In build · Repo · Site

Sentinel Engine

A customer-service agent for bank transaction disputes, in Spanish and Portuguese: the customer names a charge, and the assistant verifies the facts behind it.

  • Today: it handles the intake and triage of a dispute, verifies the facts, and hands the decision to an advisor with the full case.
  • Key decision: the LLM only labels intent; masking, policy, confirmation, case creation and handoff run in code and are verified.
  • v1.1.0 · active · Repo · Site

Knowman

RAG over your Markdown notes: ask a question and get the passages and the exact lines behind the answer.

  • Today: it answers from one notes folder through a CLI, an HTTP API and an MCP server, and scores its retrieval with a labeled dataset on every run.
  • Key decision: a distance threshold decides what counts as evidence; when nothing qualifies, the model is never called and the result is the explicit negative.
  • v0.2.4 · Repo · Site

How I work

  • The LLM handles the language; the rules that matter live in code.
  • Nothing ships before it passes an evaluation.
  • Local-first, with verifiable evidence.

Elsewhere

Website · LinkedIn · Credly

Pinned Loading

  1. knowman knowman Public

    RAG over Markdown notes that cites file and line, or says there's no evidence. Postgres + pgvector, Ollama, MCP server, groundedness eval.

    Python 1

  2. sentinel-engine sentinel-engine Public

    Customer-service agent for bank transaction disputes: the LLM labels intent; policy, confirmation and human handoff run in code. Continues the Factored AI & Data Hackathon 2026 project.

    Python 1

  3. flow-audit flow-audit Public

    Audit workbench for BSS order flows: catalog, OMS, BPM and provisioning. Local-first CLI, no server.

    TypeScript 1