Skip to main content

// portfolio · 1 flagship + 5 systems

The work runs while you read this.

Every number below is live.

Six systems, each with a public repo, a live deploy, and a figure read straight from its endpoint right now. One is production-grade. Three are prototypes with benchmarks you can rerun yourself. Two are honest showcases. Nothing here is a screenshot.

checking the fleet…

//the fleet, mapped

One production reference at the core, five supporting systems in orbit. Node brightness tracks live endpoint health; click any node to jump to its row.

// production1

NXProduction

NexusRAG

Multi-tenant RAG platform

Multi-tenant, multi-cloud RAG platform with SSO, SCIM, RBAC/ABAC, envelope encryption, multi-region failover, and SOC 2 automation. The shipped reference every other repo points to.

retrieve → rank → generate → stream

  • LangGraph
  • FastAPI
  • pgvector
  • Bedrock + Vertex

// prototype3

DWPrototype

Data Watchtower

Pre-pipeline drift detection

Schema drift, value distribution shifts, and cardinality collapse caught at the producer boundary — before downstream pipelines break. Ships a real local profiler with markdown incident reports and a build-gate on threshold breach.

profile → compare → alert → gate

  • Python
  • pandas
  • PostgreSQL
  • Docker
EVPrototype

EvalOps

Local-first eval + regression gate

Local-first eval harness with a public, reproducible benchmark: SQuAD-style scoring across strategy variants, a pinned-baseline regression gate, and committed run history. Evaluation as a contract the build must satisfy, not a dashboard you check after.

define → run → score → gate

  • Python
  • GitHub Actions
  • Next.js
  • Vercel
RSPrototype

Revenue Signal

Explainable lead scoring

Explainable lead scoring with named-signal traces. Typed signals (job-change, hiring, infra-shift, internal notes) with recency decay; every score the rep can defend in a pipeline review without invoking a black box.

ingest → decay → score → brief

  • Python
  • FastAPI
  • pandas
  • PostgreSQL

// showcase2

ARShowcase

Agent Runbook

Durable execution

Durable execution layer for AI-assisted operational runbooks. State machine with approval gates, idempotency keys, signed audit ledger. The substrate ops automation needs before bolting an LLM on a cron script.

plan → approve → execute → audit

  • FastAPI
  • PostgreSQL
  • Celery
  • Playwright
querying endpoint…
RDShowcase

Repo RAG Debugger

Source-aware debugging

Source-aware debugging assistant. Indexes codebases, stack traces, and docs; refuses fixes that reference phantom imports, deleted-function symbols, or stale APIs. Symbol-level RAG for the post-error workflow.

index → trace → retrieve → propose

  • Python
  • FastAPI
  • SQLite
  • Embeddings
querying endpoint…

Don't take my word for it. Rerun the number.

git clone https://github.com/IgnazioDS/evalops-workbench
cd evalops-workbench && pip install -e .
python -m evalops_workbench.benchmark_runner

Every prototype ships its fixture in-repo. Clone any one and you get the same number this page is showing, offline, no credentials, in about a minute. The method is on the engineering page.

Building something that has to survive real users?

Start a conversation