Skip to content
Contact

PLATFORM

Documents answer, repetitive work runs on its own.

ARCH Platform brings the organization's dashboards, automations, reports, models, and document archive into a single system of record; questions are asked of the documents themselves, repetitive steps are bound to rules, and manual work shrinks.

5 Ürün · Tek Platform

On infrastructure in Türkiye

Enterprise SLA

COMPANY BRAIN

A company brain is built from five parts

The company brain ARCH builds brings together the dashboard a company runs its work from, the automation of that work, its reporting, its learning models, and its document archive in a single system.

These five parts are not tools sold separately. All of them are built on the same system of record: the figure shown on the dashboard, the rule that triggers the automation, and the history fed into the model are all read from the same source. When one part changes, the others see that change too.

  • 01

    Dashboard

    All of the company's work is managed from a single surface; each unit sees its own screen, and all of them look at the same record.

  • 02

    Automation

    Repetitive steps are bound to rules; the system runs the work itself, and people step in only where a decision is needed.

  • 03

    Reporting

    The record each figure is derived from is defined; the report is not a manually prepared file but the system's own output.

  • 04

    Learning models

    Models built on the organization's historical records are connected to the dashboard as decision support; forecasts do not sit in a separate tool.

  • 05

    Document intelligence

    Contracts, policies, reports, and correspondence are brought into a single search structure; questions are asked of the document itself.

Company brain — four components on a bench, dashboard on topFour components running on the same system of record — automation, reporting, learning models, and document intelligence — carry their output up to the dashboard surface along vertical lines. The dashed frame at the edge of the bench is the organization's boundary: data does not leave this frame.ORGANIZATION BOUNDARYdata never leavesDASHBOARDsingle management surfaceAUTOMATIONrule-bound executionREPORTINGfigure derived from the recordLEARNING MODELSforecast from historyDOCUMENT INTELLIGENCEsearch and sourceDIAGRAM · COMPANY BRAIN
Four components run the work on the same bench; the dashboard on top forms a single view. Each component carries its own output up to the dashboard. The dashed frame is the organization's boundary — data does not leave that frame.

Four components run on a single system of record, and their outputs come together on the single dashboard surface above. Components:

  • Automation: rule-bound execution
  • Reporting: figure derived from the record
  • Learning models: forecast from history
  • Document intelligence: search and source

Dashboard: single management surface

Organization boundary: data never leaves

DOCUMENT INTELLIGENCE

Questions are asked of the document itself

The organization's documents stop being a readable archive and become a source you can ask questions of.

The organization's contracts, policies, technical reports, and correspondence are brought into the system. When a question is asked, the system first finds the relevant sections within these documents, writes the answer based solely on those sections, and shows which part of which document it came from. The source sits beneath the answer; opening the document is enough to verify it.

Search does not depend on exact word matches. When you ask "how do I terminate the contract," it also finds the clause governing notice of termination even if that clause never contains your sentence. The person asking does not need to know the term used in the document.

The knowledge boundary is the organization's own document set. The model does not use general knowledge outside that set as a source, the deployment runs on the organization's own hardware, and a document a user cannot see does not appear in the answer given to that user either.

MACHINE LEARNING

A forecast that feeds a critical decision must be validated

ARCH's forecasting models are trained on the organization's own historical records and run on the organization's own hardware. The same discipline applies to every model that feeds a critical decision: the data it was trained on is recorded, its behavior over past periods is measured, and its drift in production is monitored.

  • Training on the organization's own data — rather than asking a general-purpose model, a model is built on the organization's records.
  • Backtesting — the model is run over a past period, and the forecast is compared with actuals.
  • Probability range — a range is given rather than a single number, so it is visible how much uncertainty a decision is made under.
  • Drift monitoring — when the error in production grows, an alert is raised; the model does not degrade silently.
  • On-premises operation — training and inference on the organization's hardware; data never leaves.

01 · ARCHITECTURE

Production-grade infrastructure.

Built on AWS EC2, Coolify, and Postgres; each product is deployed as an isolated service, and the platform layer is shared.
ARCH Platform — four-floor shared infrastructureFour floors, bottom to top: data ingestion, model layer, orchestration, and observability. Five ARCH products sit on either side of the stack and connect to the floors they consume.REST APIjson · webhookSFTPbatch · ackTAPU / TKGMe-devletOPEN-METEOweather · 10 minPOSTGRESoltp · replicaSENSORmqtt · scadaKAFKAevent · logS3 / OBJECTmedia · pdfLLM INFERENCEprompt · toolCLASSIFICATIONtabular · xgbFORECASTforecast · 168hNLP PIPELINEner · summarySTTvoicetriggernormalizevalidationinferencedispatchpublishLATENCY84msERROR0.12%DRIFT0.04 σUPTIME99.96%01DATA INGESTION02MODEL LAYER03ORCHESTRATION04OBSERVABILITYBORAReal estate valuation operating systemPRODUCTCallCenterAICall center automationPRODUCTWindSight168-hour wind energy forecasting enginePRODUCTTakbisTAPU-TKGM query and check networkPRODUCTGabimAutomation129-field valuation form extractionPRODUCTSINGLE RUNTIME5 products · shared infrastructureON INFRASTRUCTURE IN TÜRKIYEAWS EC2 + CoolifyKVKK COMPLIANTPostgres + Grafana · auditableARCH PLATFORM · v1.0

Four floors of shared infrastructure, ordered bottom to top. Around each floor, the ARCH products that consume it are listed.

  1. 01 — DATA INGESTION
  2. 02 — MODEL LAYER
  3. 03 — ORCHESTRATION
  4. 04 — OBSERVABILITY
  • BORA: Real estate valuation operating system
  • CallCenterAI: Call center automation
  • WindSight: 168-hour wind energy forecasting engine
  • Takbis: TAPU-TKGM query and check network
  • GabimAutomation: 129-field valuation form extraction

02 · HOW IT WORKS

Data goes in. Decisions come out.

  1. Data Connection

    The platform automatically collects data from public records and from the organization's own systems.

  2. Normalization

    Raw data is converted to a common schema; missing fields, outliers, and format mismatches are resolved automatically.

  3. Model Inference

    A product-specific ML/LLM model runs; the result returns with a confidence score and an explanation label.

  4. Trigger Workflow

    Output is passed downstream automatically (CRM update, SMS, report PDF, voice script).

  5. Monitoring & Alerts

    Every step is logged; when thresholds are exceeded, a Grafana alert and an operator notification are triggered.

Data ingestion — from scattered sources to a single system of recordSix data sources scattered around the edge — title deed record, bank valuation request, call record, weather and wind data, leave request, vehicle location — flow along cable lines into the single decision center in the middle.DECISION CENTERsingle system of recordTITLE DEED RECORDTAPU · TKGMWEATHER AND WIND DATAOpen-MeteoVALUATION REQUESTbank workflowLEAVE REQUESTpersonnel recordCALL RECORDvoice customer lineVEHICLE LOCATIONfleet recordDIAGRAM · DATA INGESTION
From scattered sources to a single system of record: each line carries the record at its source to the decision center in the middle.

Scattered data sources flow into the single system of record in the middle. Sources:

  • Title deed record: TAPU · TKGM
  • Valuation request: bank workflow
  • Call record: voice customer line
  • Weather and wind data: Open-Meteo
  • Leave request: personnel record
  • Vehicle location: fleet record

Decision center: single system of record

PLATFORM METRICS

  • 5

    products live in production

Products running on the platform

All products (5)

Let's talk about your system

Tell us which process you want to automate; we will draw up an actionable plan.

Get in touch