● Independent AI engineering practiceEnergy · Manufacturing · Industrial IoTPrague · working across DE, AT, CZ
AI that runs in the plant.Not in the slide deck.
ioxoi is the independent practice of Homan Ansari. I build and run production AI for energy and industrial companies: vertical agents inside your workflows, IoT platforms that predict failures, and the leadership to make both stick. Compliance built in from day one, measured after go-live.
Agents in production8HR · IT · Sales · Finance · PR · Ops · Support · Voice
Failure lead time72hwas: minutes, from SCADA codes
Maintenance cost−40%12 months of fleet operation
Fastest go-live3 dayscompany chatbot, concept to production
01 — Services
Four things I do. All of them end in production.
No labs, no pilots that die in PowerPoint. Every engagement is scoped to a workflow, shipped into your systems, and measured against a baseline we agree on first.
S-018 in production
Vertical AI agents, built and run
Agents for the workflows you already have: support, IT help desk, sales qualification, invoice processing, site operations. Built on your data, deployed inside your compliance boundary, monitored after go-live.
RAG
n8n
Claude · GPT · local models
first agent live in days
S-02100+ turbines live
Industrial IoT & predictive maintenance
Edge-to-cloud platforms for turbines, plants and fleets. Sensor ingestion across hardware generations, time-series storage, anomaly detection and failure forecasting, with dashboards your engineers actually use.
AWS IoT Core
InfluxDB · Grafana
PyTorch · LSTM
Isolation Forest · anomaly detection
S-031–2 days / week
Fractional AI leadership
Head of AI on demand for mid-size industrial companies: strategy, build-vs-buy, vendor selection, hiring, delivery cadence. Hands on the code when the team needs it, in the boardroom when the plan needs it.
30 years in IT
Director of Software Engineering & IT
German (native)
S-04built for audit
AI security & compliance
Architectures and reviews built for audit: GDPR data residency, ISO 27001-aligned controls, SOC 2-style logging, threat modelling for models, prompts and agent tool access.
ex-HPE Enterprise Security
CEH · CISSP & CISM training
ISO 27001-aligned
02 — Work
Agent register
Built and run by me
IDAgentOutcomeStatusStack
01Wind energy IoT platformPredictive maintenance for 100+ turbines, −40% maintenance cost, 72 h lead time● LIVEAWS IoT · InfluxDB · PyTorch
02Operations agentReal-time KPI tracking, anomaly detection, incident reports and escalation for industrial sites● LIVEIoT · alerting · automation
03Company chatbot−60% support queries, 95%+ accuracy over 10,000+ documents, live in 3 days● LIVERAG · Pinecone · LangChain
04IT & Dev agentTier-1 tickets handled autonomously, −50% resolution time, on-demand code generation● LIVEClaude Code · GitHub
05Sales agent3× faster lead response, proposals generated in minutes● LIVEGPT · CRM · n8n
08PR agentPress releases drafted in minutes, 24/7 media and brand sentiment monitoring● LIVEClaude · sentiment
09Voice assistantVoice assistant for calendar and email automation, 10+ hours saved weekly● LIVEElevenLabs · Vapi
Agents built and run in my employed role and in my own operations. Figures are measured against each system’s pre-launch baseline; method on request.
03 — Case study · wind fleet
From SCADA error codes to 72 hours of warning
Built and run in my employed role leading software engineering and IT at a wind energy company; shown as a reference, not as an ioxoi client project. 100+ modular turbines, reactive maintenance, data in silos. Deployed across the full fleet in 4 months, measured over 12 months of production.
Layer 1 · EdgeAgents on every controller
40+ sensor channels sampled under 100 ms. Local pre-filtering ships only deviations: −70% bandwidth.
→
Layer 2 · StreamOne pipeline, three hardware generations
AWS IoT Core ingestion, InfluxDB time series, a normalisation layer across incompatible protocols.
→
Layer 3 · IntelligenceTrain on the 72 h before failure
Isolation Forest for anomalies, LSTM for forecasting. Failures are rare; the signature before them is not.
−40%maintenance cost
72hprediction lead time
12months in production
−70%bandwidth via edge
The hard parts
Data quality.Intermittent connectivity, sensor drift and firmware bugs inject corrupt readings. A validation pipeline separates “sensor failure” from “real anomaly”; getting it wrong means constant false alarms or missed failures.
Compliance.GDPR data residency, audit trails and change management, with ISO 27001-aligned controls planned in from the start.
04 — Approach
How an engagement runs
Fixed-price project, monthly retainer or fractional leadership at one to two days a week. Rates on request. Nothing is locked in: your team can take over at any point.
STEP 1 · TWO DAYSDiagnose
Two days with your people, on site or remote. I map the workflow, the data, the systems and the compliance boundary. You get a written plan: scope, risks, baseline metric and a fixed price for the first agent.
STEP 2 · DAYS TO WEEKSShip
The first agent goes live in production with real users and real data. Days for a contained workflow, a few weeks when legacy integrations are involved. Measured against the baseline from step 1.
STEP 3 · MONTHLYOperate
Monitoring, evaluation, model updates and a monthly report. Handover to your team whenever you want it, with documentation and runbooks that survive without me.
05 — Behind ioxoi
[PORTRAIT]
Homan Ansari
I have run infrastructure since 1996, founded ioxoi in Berlin in 2011, spent years in enterprise security at Hewlett Packard Enterprise, and today lead software engineering and IT at MOVEAIR, a wind energy company. ioxoi runs alongside that role and makes this experience available to other industrial companies. You work with me directly; there is no junior team behind the pitch.
Not a mock. A live agent that answers questions about the services, the case study and how an engagement works. It is also the kind of thing I build for clients.
ioxoi agentlive
Hello. I'm the ioxoi agent. Ask me what a first engagement looks like, how the turbine platform was built, or whether your workflow is a fit.
AI assistant, not a human. Your messages are processed by AI service providers to answer you and deleted after 14 days. Please don't enter confidential data.
·
07 — Contact
Bring me a workflow that hurts.
One email with the workflow, the systems involved and who owns it. I reply within one business day with whether it's a fit and what the two-day diagnosis would cost.
