Gabriel Valerio · Automation & AI Systems

Systems that work while you sleep.

I design, build, and operate automations for teams drowning in manual work — connecting tools, automating decisions, and turning raw data into action. The event feed shows real output from systems I run in production.

gabriel@valeriodata.com View work ↓

50+workflows in production

6industries served

24/7monitored & alerting

LIVE · SYSTEMS RUNNING

ANONYMIZED

EVENTS FROM CLIENT SYSTEMS IN PRODUCTION · DETAILS REDACTED

Selected systems

10 / 10 SHOWN

Real production systems — client work and flagship builds. Company details anonymized where needed; numbers are real.

CLIENT: SPORTS-TECH SAAS (US) 2026 — NOW IN PRODUCTION

Hybrid RAG + knowledge-graph context engine

Local retrieval engine that gives AI agents grounded knowledge of a client's entire ops stack — Salesforce flows, playbooks, precedent cases, meeting transcripts. Four fused strategies: semantic embeddings, BM25 keyword, entity-graph traversal, and recency/authority weighting. Runs entirely on a local GPU — no per-query API cost, no data leaving the machine.

STACK: PYTHON · SQLITE-VEC · FTS5 / BM25 · QWEN3 EMBEDDINGS · ENTITY GRAPH

1,018knowledge units indexed

1,477entity graph edges

12s → 0.13squery latency, cached

CLIENT: SPORTS-TECH SAAS (US) 2026 — NOW IN PRODUCTION

Production AI support agent

An AI helpdesk agent that auto-resolves the routine slice of inbound support across live chat and email, so the team absorbs volume without hiring. Deterministic 5-stage RAG grounded on a curated knowledge base, with retrieval-confidence gating, a hard always-escalate list for billing and legal, and an independent second-model verifier that downgrades any ungrounded answer to a human. US-pinned model hosting for FERPA-scoped data.

STACK: NODE / EXPRESS · SUPABASE · RAG · OPENROUTER · HMAC WEBHOOKS

99%correct on 162-case eval

0hallucinations · 100% grounded

2channels · chat + email

CLIENT: ENTERPRISE RETAIL 2026 DELIVERED

Bidirectional Monday ↔ Azure DevOps sync

A 7-scenario iPaaS pipeline keeping a retail giant's Monday.com boards and Azure DevOps backlog in lockstep — create, update, and close Epics and Features in both directions with loop-safe conditional writes across an 8-field mapping contract spanning two very different API surfaces.

STACK: MAKE.COM · MONDAY GRAPHQL v2 · AZURE DEVOPS API

~90%manual sync ops cut / sprint

<5swebhook-to-write latency

7scenarios · bidirectional

CLIENT: COMMODITIES (LATAM) 2026 DELIVERED

iPaaS cost audit + zero-disruption migration

Root-caused a runaway automation bill — an unscoped trigger firing on every board event, a retry storm spiking 40 concurrent runs in 400ms, and state-machine logic paid for in iPaaS that the platform handles natively for free. Rebuilt 25 native automations behind a parallel-run parity check and cut over with zero downtime.

STACK: MAKE.COM · MONDAY.COM NATIVE · NODE / EXPRESS POC

39,679/mooperations reclaimed

25automations rebuilt native

0pipeline disruptions

CLIENT: INVESTMENT MANAGEMENT 2026 DELIVERED

KYC / AML compliance automation

Automated counterparty screening for an investment manager — LSEG World-Check One and Netrin checks with structured PDF dossiers written back into Monday.com — plus HR admission and staff-movement email automations. Debugged entirely in production against no staging, behind a standardized backup → validate → verify → re-activate deploy protocol.

STACK: MAKE.COM · LSEG WORLD-CHECK ONE · NETRIN · MONDAY · DOCUSIGN

3silent go-live bugs caught

0false positives after fix

10+automations delivered

CLIENT: EDTECH (US) 2025 DELIVERED

Monday.com → Jira Cloud migration engine

Production-grade Python ETL that moved an entire EdTech company off Monday.com. Wave-based orchestration with state management, rollback handlers, dry-run mode, and post-migration validation — so the team kept working while the ground moved under them.

STACK: PYTHON · MONDAY API · JIRA CLOUD · ETL

14,000+items migrated

0records lost

3 wk → 4 dvs. manual plan

FLAGSHIP BUILD 2026 — NOW IN PRODUCTION

Ambient-intelligence platform + autonomous agent

An always-on "second brain" that captures on-screen work and meetings as searchable text, OCRs every frame and deletes the image immediately — keeping only text — then layers a human-gated AI agent that mines action items and drafts responses. Nothing outward-facing acts without approval. Semantic search over local embeddings with hybrid BM25 × confidence × recency re-ranking.

STACK: PYTHON · SQLITE-VEC / FTS5 · WHISPERX · OLLAMA · 22 SYSTEMD SERVICES

~800×storage reduction vs raw

61% → 100%retrieval hit@10, tuned

22services · self-healing

DATA PLATFORM 2026 — NOW DEPLOYED · MAINTAINED

Electoral-intelligence analytics platform

A full-stack analytics platform that drills election data from state down to the individual voting location, with choropleth and heatmap geospatial views and a campaign field-team management console. Ingests and models Brazil's open government datasets in BigQuery, with real entity-matching to align voting sections to neighborhood polygons.

STACK: REACT 19 · DECK.GL / MAPBOX · FASTAPI · GOOGLE BIGQUERY · DOCKER / VPS

31municipalities modeled

4drill levels · state → location

8analytical views

CLIENT: CONTENT CREATOR 2025 — NOW IN PRODUCTION

Social media content factory

AI-driven clip pipeline: Whisper transcription, controversy scoring via keyword heatmaps, viral-moment selection, automated ffmpeg cutting, and caption generation for TikTok, Reels, and Shorts. Runs end-to-end without a human in the loop.

STACK: CLAUDE CODE · WHISPER · FFMPEG · PYTHON

36/wkclips produced

~15 h/wkediting eliminated

PERSONAL SAFETY SYSTEM 2024 — NOW IN PRODUCTION

Pool safety monitoring with AI detection

24/7 camera monitoring with motion detection and person recognition. Multi-channel alerting — Telegram, email, and on-site audio — fires the moment someone approaches the water. Fully autonomous since deployment.

STACK: PYTHON · TUYA CLOUD API · TELEGRAM BOT · AI DETECTION

<2salert latency

24/7autonomous uptime

Capabilities

04 AREAS

Code when it needs code, no-code when it doesn't. The right tool is the one your team can live with.

Workflow automation

End-to-end process automation with Make.com, n8n, Zapier, and GitHub Actions — from simple triggers to multi-step orchestrations.

Data & analytics

Power BI dashboards, Python pipelines, BigQuery analytics, and statistical models that turn raw data into decisions.

AI integration

LLM agents, MCP servers, and Claude / OpenAI / Grok implementations wired into real business processes.

No-code systems

Airtable and Monday.com platforms your team can actually operate and extend after I hand them over.

Have a workflow problem?

There's probably an automated solution. Tell me what your team does by hand and I'll tell you what a system could do instead.

Send email LinkedIn