Skip to content
Sachin Rana

Agentic AI Architect · Fractional CTO

AI agents that solve real problems, on systems you own.

I design and ship agentic AI systems end to end: the agents, the data pipelines that feed them, the digital twins they reason over, and the infrastructure they run on. One accountable architect, from sensor to decision.

Send a project briefTry the agent demo

Sachin RanaFounder, Ideaxa Solutions · New Delhi, India

● Edge → data → agent → action

How an agentic system turns field data into actionField sensors publish through an edge gateway and MQTT into ClickHouse. An AI agent queries the data, reasons with a private LLM gateway and the digital twin, passes a human approval gate, then raises an action such as a work order.Field sensorsmeters · PLCs · BMSEdge gatewayModbus · BACnetMQTT brokerEMQX clusterClickHousesub-50 ms queriesDigital twinphysics healthAI agentplan · MCP tools · memoryqueryhealthreasonLLM gatewayprivate GPUsHuman approvalguardrail gateActionwork order · PR · alert
ExampleCondenser approach drifting → agent checks the twin → work order drafted for approval
Cloud spend removed through bare-metal repatriation
85%Cloud spend removed through bare-metal repatriation
Reliability on mission-critical SCADA telemetry
99.999%Reliability on mission-critical SCADA telemetry
Production physics-informed digital twins
9Production physics-informed digital twins
Years shipping production systems
10+Years shipping production systems

01Agentic AI

Pick a problem. Watch the agent work it.

This is how my agent systems reason through a real request: plan, call tools against live data, stop at a guardrail for a person to approve, then act. Each walkthrough maps to delivered work.

Agent trace · simulated walkthrough

Goal

Find what is driving spend and propose a cut that keeps latency and security intact.

    A scripted walkthrough of how my agent systems reason. No live model runs here, and nothing you click leaves your browser.

    How I build agents: gateway, tools, memory, guardrails

    02Digital twins

    Not a dashboard. A live model that finds faults.

    Launch a twin, inject a fault, and watch the health engine catch it and price the energy lost. These run fully in your browser on simulated plant data.
    Open full screen

    Live digital twin

    AI Pod · liquid-cooled GPU racks

    Runs fully in your browser · about 1–2 MB

    Four 80 kW GPU racks on direct-to-chip cooling, a CDU and a roof dry cooler. The twin judges every loop against healthy physics and prices the energy lost.

    Try this

    1. Press “Cold-plate blockage, rack 3” under Inject a fault
    2. Watch rack 3’s flow halve and its return run hot, red within seconds
    3. Acknowledge the alarm, then repair and watch it return to normal
    Get a twin of your plant

    Built by Ideaxa Solutions with simulated, generic plant data. Isometric drawing kit by Ryan, used under the MIT license.

    All twins and the nine production platforms

    03Selected work

    Outcomes clients paid for

    Client names are withheld under NDA. Figures describe the delivered production systems.

    01 · Cloud cost

    85% Cloud Cost Reduction via Bare-Metal Hybrid Migration

    High-Growth AI & Analytics SaaS

    • 85% Cost Savings
    • <45ms Latency
    • 0 Unencrypted Paths
    Challenge
    Client faced unsustainable 6-figure monthly cloud bills on managed databases (RDS, vector search) and LLM API tokens, with query latency exceeding 350ms.
    Approach
    Architected a hybrid bare-metal repatriation strategy. Migrated primary analytical storage to a self-hosted ClickHouse cluster and deployed private GPU nodes for local model inference, connected back to AWS via NetBird zero-trust WireGuard mesh.
    Outcome
    Reduced recurring cloud infrastructure spend by 85%, reduced query latency from 350ms to <45ms, and established 100% data sovereignty.

    02 · Cloud security

    Zero-Trust Enterprise Multi-Account Cloud Migration

    Regulated Financial Services Provider

    • SOC 2 Pass
    • 100% Automated IaC
    • Zero Ingress Leaks
    Challenge
    Legacy monolithic AWS infrastructure with shared credentials, lack of network micro-segmentation, and impending SOC 2 Type II audit deadlines.
    Approach
    Engineered modular Terraform landing zones with AWS Control Tower, SCP boundaries, Transit Gateway hub-spoke topology, and NetBird zero-trust access control.
    Outcome
    Achieved 100% automated infrastructure provisioning, passed SOC 2 Type II audit with zero findings on initial evaluation, and eliminated all public IP exposures.

    03 · Industrial telemetry

    Mission-Critical SCADA & Smart Grid Telemetry Modernization

    Commercial Energy & Facility Operator

    • 99.999% Reliability
    • Sub-50ms Forensics
    • 0 Rollover Skews
    Challenge
    Packet loss across remote industrial meters and erroneous power consumption metrics caused by trapezoidal integration rollover errors.
    Approach
    Deployed industrial LoRaWAN and Modbus RTU gateways running hardened C++/Python firmware with local failover buffering. Ingested raw telemetry into ClickHouse with deterministic MAX - MIN hardware register delta formulas.
    Outcome
    Eliminated accumulator rollover calculation errors, achieved 99.999% telemetry reliability, and unlocked real-time forensic detection of voltage sags.
    All case studies and animated reference architectures

    04End to end

    Four disciplines, one accountable architect

    Clients come in for one and stay because they connect. Telemetry feeds the twins, the twins feed the agents, and all of it runs on infrastructure you control.

    Agentic AI & LLMOps

    Agents that reason, call tools and ask before they act.

    Architecting sovereign on-premise AI inference stacks, centralized LLM routing proxies, FastMCP tool ecosystems, and multi-agent autonomous engineering teams with episodic memory.

    • LiteLLM
    • vLLM
    • Ollama
    • FastMCP
    • MemOS
    • Qdrant

    Digital Twins

    Live models of real plant that find faults on their own.

    Operating 9 production-grade, physics-informed digital twins combining dynamic simulation with physical ClickHouse hardware telemetry.

    • Next.js 14+
    • SVG Vector Modeling
    • WebGL
    • Server-Sent Events
    • Physics Modeling
    • Docker

    Industrial IoT & Telemetry

    Field data captured without loss and queried in milliseconds.

    Architecting ultra-reliable industrial telemetry backbones, high-speed ClickHouse time-series data lakes, enterprise MQTT clustering, and forensic power quality analytics.

    • ClickHouse
    • EMQX Enterprise
    • Node-RED
    • LoRaWAN
    • Modbus RTU/TCP
    • BACnet

    Cloud & Infrastructure

    Platforms you own, secured and costed like a CFO would want.

    Architecting robust enterprise cloud landing zones, automated Terraform IaC pipelines, high-density bare-metal hypervisors, and autonomous zero-cloud-tax PaaS deployment platforms.

    • Terraform
    • AWS
    • Azure
    • Proxmox VE
    • Docker
    • NetBird
    • Amazon Web Services
    • Microsoft Azure
    • Terraform / IaC
    • Proxmox VE 8+
    • Linux Kernel & Systems Administration
    • Docker & Docker Compose
    • LiteLLM Gateway
    • Local Inference Runtimes
    • Model Context Protocol
    • Multi-Agent Systems & A2A Agency
    • ClickHouse Time-Series DB
    • DuckDB & Parquet Analytics
    • Industrial Modbus & BACnet
    • LoRa / LoRaWAN
    • ESP32 & Microcontroller Firmware
    • NetBird & WireGuard Mesh
    • CrowdSec & Network Defense
    • Python
    • Node.js & Next.js
    • C / C++
    • Microsoft SQL Server
    • PostgreSQL & Redis
    • Qdrant Vector Database

    05Governance

    Built to pass audit, not just demo

    Agents and platforms are designed against twelve regulatory and security frameworks, from SOC 2 and HIPAA to IEC 62443 and the EU AI Act. Each maps to concrete controls.
    • SOC 2 Type II
    • ISO/IEC 27001:2022
    • HIPAA / HITECH
    • GDPR & India DPDP Act 2023
    • NIST SP 800-53 Rev. 5
    • PCI-DSS v4.0
    • CIS Benchmarks Level 1 & 2
    • IEC 62443-4-2 / ISA-99
    • FedRAMP High / Moderate
    • ISO 9001:2015
    • NIST CSF 2.0
    • OWASP Top 10 & LLM Top 10
    See the control evidence for each framework

    Got a problem an agent, a twin or a better platform could solve?

    Send a two-minute brief. You get a reply from me with a first view on approach, timeline and cost.

    Send a project briefEmail directly