Horizon forecasts demand for retail and manufacturing. Pliney resolves operations requests with cited sources. When the problem doesn't fit a SaaS, we engineer the system end to end.
30-minute intro call. We'll never share your email. Privacy policy
From global brands to research institutions. Sixteen teams shipping demand forecasting, grounded agents, and bespoke AI systems in production today.
Most AI projects die between the demo and production. Ours don't.
Every Horizon deployment, every Pliney agent, every custom system we engineer runs with citations on every answer, evaluation in CI, real ERP and helpdesk connectors, and tenant data that never crosses a wall. AI that survives contact with reality.
Two SaaS products and a custom-AI engineering practice. Each one shipped, each one observable, each one running for real customers today.
Forecast, simulate, replenish, execute. Multi-level, channel-aware, integrated with the ERPs and commerce stacks your planners already run on.
RAG over your sources, multi-agent routing, live tool use over your ERP and helpdesk. Forecasts are queryable in plain language via Horizon.
Discovery sprint, fixed-scope pilot, production rollout. Agent pipelines, RAG systems, data platforms, built on the same engineering principles that run our products.
An incubator on the ground in Porto and a hyperscaler underwriting our infrastructure. The institutions our customers expect to see standing beside us.
Horizon's operating loop, illustrated end to end: from a demand-shift detection to an autonomous reorder. Pliney runs the same loop on the operations side. Custom AI Solutions adapt it to whatever shape your problem takes.
Sofia CostaHead of Buying & Merchandising
Horizon
Active now · with Sofia
Sofia, Linen Shirts demand is up 22%. At this pace, you'll stock out in 6 days.
9:42 AMWhich stores are most at risk?
9:44 AMStores 3 and 7. Both below 4 days of coverage.
9:46 AMDetect
Demand shifts, stock imbalances, and sell-through anomalies are continuously monitored, so your team knows what's changing before it's too late.
Ricardo MatosOperations Manager
Horizon
Active now · with Ricardo
Priority for today: transfer 90 units of Basic Tees to Store 3. €6.2K at risk.
9:49 AMApprove.
9:51 AMPresent
Each role sees the next decision to make, ranked by urgency and impact. Not another dashboard to interpret.
Marta FerreiraFinance Director
Horizon
Active now · with Marta
Anything under €5K, just go ahead. Flag me on the bigger ones.
9:56 AMUnderstood. Routine reorders will execute automatically within your rules.
9:58 AMGuide
Your team defines the rules: budget thresholds, quantity limits, approval flows. Horizon operates within those boundaries and escalates when it should.
Joana AlmeidaPlanning Manager
Horizon
Active now · with Joana
Good morning. Overnight: 12 reorders placed, 4 transfers executed. 1 item flagged for your review.
10:03 AMShow me the flagged one.
10:05 AMExecute
Routine decisions happen automatically: reorders, transfers, alerts. Your team wakes up to results, not to-do lists.
Miguel SantosCOO
Horizon
Active now · with Miguel
How did we perform this week?
10:10 AM€47K revenue protected. 23 decisions executed autonomously. 2 required your team's input.
10:12 AMYour team spends less time on routine decisions every week, and that gap keeps growing.
10:14 AMLearn
Every decision makes the next one better. Over time, Horizon needs less input and delivers more impact.
One loop, one product. Horizon turns routine demand decisions into autonomous, auditable execution inside rules your team sets.
Reorder Linen Shirt
Stockout risk in 6 days
€18.4K at risk
Buying
Approve reorder of 240 units
Finance
Validate PO impact and budget
Marketing
Reduce campaign until stock arrives
Operations
Rebalance stock to top location
Different teams see different actions, all from the same shared decision layer that powers Horizon, Pliney, and the custom systems we ship.
Pliney isn't a router. It's a grounded agent that goes all the way to a resolution. Every reply cites its source. Every action runs over real ERP, CRM, and helpdesk connectors. RAGAS scores every turn in the loop.
A customer or operator pings the channel they already use: email, Slack, helpdesk. Pliney picks up the request, classifies intent, and decides whether to ground in docs, query live data, or both.
Pliney calls the ERP, the WMS, the helpdesk's own API, whatever it takes, over MCP and REST. Not a frozen knowledge dump: the same source of truth your operators are watching.
Every resolved conversation is scored on its way out: grounded against the sources Pliney actually pulled, judged for faithfulness, relevancy, and recall, and stored for audit. Nothing ships without RAGAS thresholds met.
Where's my order #2231? It was supposed to ship Tuesday.
Turn 1 · inbound · HelpdeskConfirmed shipment from WH-2, DPD GE2231887, ETA Thu 9 to 11.
Turn 2 · grounded answerGreat. Can you push the ETA to the customer over WhatsApp?
Turn 3 · inbound · HelpdeskWhatsApp sent with tracking, ticket marked resolved, CSAT scheduled Friday.
Turn 4 · action takenThe same engineering principles drive Horizon, Pliney, and every Custom AI Solution we build for clients. AI that ships is AI you can trace, evaluate, and govern.
Every answer cites its source. RAG pipelines you can audit, with citations surfaced to the user.
RAGAS auto-evaluation on every turn. Structured tracing, span-level visibility, evaluation in CI.
REST, GraphQL, ERPs, helpdesks, MCP servers. Real integrations, not webhook stubs.
Multi-tenant boundaries enforced at the data model. Tenant-scoped models, raw payload preservation for audit.
About a third of what ships from ZAAI is bespoke. Industries we serve at depth, paired with the production architectures we've built, so you can see both the demand and the proof.
Thousands of SKUs across hundreds of stores. Promotion-aware demand forecasting, replenishment, and waste reduction at scale.
Multi-stage production planning, raw-material forecasting, capacity simulation, wired directly into the ERP and MES.
Seasonal demand, style-level forecasting, and store-by-store replenishment for collections with short shelf lives.
Grounded agents over regulated data, compliance gap analysis, and decision support inside the firewalls finance teams require.
Multi-stage agent pipeline reviews regulatory documents and emits structured gap reports. Quarterly review compressed from 3+ weeks to under 4 hours.
Elastic clustering compute on ECS Fargate with Redshift Serverless for warehousing. Compute cost down ~60% vs. always-on baseline.
Surfaces discrepancies across marketplace, OMS and ERP: returns not credited, invoiced not fulfilled, value mismatches. Reconciliation went from monthly to daily.
FORGE framework: combining foundational time-series models with classical statistical baselines. Productionized inside Horizon, built on research published at AAAI / KDD.
Founder of the AI ventures Nixar, TUTAI, and Liftter. VP of Data & AI at Marley Spoon. Co-founder and CEO of HUUB, acquired by Maersk. Author of the "Deep Learning for Time Series Cookbook" (Packt). Published at AAAI and KDD, two of the top AI research conferences worldwide. PhD researcher at the AI Lab of the University of Porto.
LinkedInTell us about the decision you're trying to make. We'll tell you whether Horizon, Pliney, or a Custom AI Solution fits, or whether it's none of the above.
Tell us what you are working on. We reply within one business day.
Candal Park, Fração AL-28, Alameda da Empresa 64, 4400-133 Vila Nova de Gaia
UPTEC Mar, Avenida da Liberdade s/n, 4450-718 Leça da Palmeira
