Cela Consulting

Financial data consulting & software development

Data you can trust. Software you can build on.

We automate financial analysis, risk management and regulatory reporting, so every number you report is checked, reconciled and traceable to its source. Our software studio builds products that last: a website builder, a construction platform that takes a building from first drawing to permit, and AI agents calibrated on your own data.

  • ConsultingFinancial analysis, risk, regulatory reporting, dashboards and data flows
  • SoftwareWebsite builder, construction software and AI agents

What we do

One firm, two practices.

Our consultants and developers work from the same methods: clean data models, automated tests and documentation that an auditor or a new colleague can follow.

Consulting

Financial data, risk and reporting

We automate how financial data is collected, checked, analysed and reported, so your team spends its time on decisions instead of spreadsheets.

  • Automated financial analysis
  • Risk management
  • Regulatory reporting
  • Dashboards and data visualisation
  • End-to-end data flows
  • Software development
Explore consulting

Software

Websites, construction software and AI agents

We design, build and run our own software products, with the same engineering discipline we bring to financial data.

  • Website builderProfessional websites your team can edit without writing code.
  • Construction softwareDraw the building, check feasibility, get the bill of materials, permit draft, statement of work and schedule.
  • AI agentsAgents that read images, locations and documents and return checked answers, calibrated on your own cases.
Explore software

Consulting

We automate the numbers behind your decisions.

From the general ledger to the regulator, we build the data flows, checks and models that produce your reports automatically, with every figure traceable to its source.

Automated financial analysis

Scheduled pipelines load, check and analyse your financial data before the working day starts. Errors are fixed at the source, so the report your team opens in the morning is already validated.

month_end_credit_report Done in 41 s · next run tomorrow 06:00
  1. 06:00:02loadloan data · 3 sources · 2,184 rows
  2. 06:00:09check312 validation rules · 5 issues
  3. 06:00:16fix5 issues corrected at source
  4. 06:00:24matchtotal to general ledger · Δ 0.0
  5. 06:00:33analyseexposure by rating and variances
  6. 06:00:41publishcredit risk report
6 / 6 steps

From raw data to a validated report

LoanCounterpartyEUR kMaturityRating
L-1042Arden Logistics NV1,250.02028-06-30BBB
L-1042Arden Logistics NVDQ-071,250.02028-06-30BBB
L-1077Brightwater Foods SA880.52027-11-15–BB+
L-1103Corvel Medical Inc1,420.0USD1,310.02029-03-31A
L-1120Delmare Shipping BV640.015/03/20272027-03-15BB
L-1138Elmstone Retail SA2,105.32030-09-30BBB
Total

Regulatory reporting

We automate prudential, resolution and DORA reporting. Every submission is checked against the regulator's own validation rules before it leaves the bank, and each error points to the record that caused it.

  • COREP
  • FINREP
  • LCR · NSFR
  • SRB
  • DORA

6.5%of DORA registers passed every data quality check in the ESAs' 2024 dry run.

    Ready to submit

    Risk management

    Models for credit, market and liquidity risk, with limit monitoring and stress tests your risk committee can follow line by line. Drag the VaR line, change the confidence level or switch to a stressed period.

    Daily P&L in millions · 1,000 trading daysBeyond VaRExpected shortfall
    VaR 95.0%
    −1.40
    Expected shortfall
    −2.10
    Breaches, last 250 days
    15expected 12.5

    Software development

    Internal tools, APIs and integrations around your finance systems, in maintainable code that you own.

    pipeline.py48 tests passed

    End-to-end data flows

    One governed flow from source systems to final reports, with lineage for every figure and alerts when something breaks.

    We design the flow around the systems you already have, and adapt it when they change.

    Drag the blocks to rearrange the flow. Hover or focus a block to trace its lineage.

    ERP
    Core banking
    Market data
    Spreadsheets
    Ingest
    Validate
    Model
    Regulatory returns
    Risk reports
    Dashboards

    Figures, records and code on this page are illustrative.

    Dashboarding

    Dashboards your team will actually use.

    We build interactive dashboards on the same governed data as your reports, so every chart traces back to a reconciled number. Swipe through five examples. Each one runs on sample data: change the filters, hover any chart, or switch a chart to a table.

    1 / 5

    Credit portfolio

    Sample data · as of 30 Sep 2026 · bn = billions, m = millions

    Period
    Segment
    Scenario

    Exposure

    Expected loss

    Non-performing loans

    Limit utilisation

    of approved limits · watch level 85%

    Exposure by segment

    bn, month end · dashed = forecast to Dec 2026

    Exposure mix

    share of exposure, Sep 2026 · click to filter

    Exposure by rating and maturity

    bn · current portfolio

    Largest exposures

    % of Tier 1 capital · limit 25%

    Liquidity and funding

    Sample data · as of 30 Sep 2026 · bn = billions

    Stress scenario

    LCR

    minimum 100% · target 120%

    NSFR

    minimum 100% · target 120%

    Liquid asset buffer (HQLA)

    Survival horizon

    days until the buffer runs out · appetite 90 days

    Maturity ladder

    contractual cash flows, bn · base

    Funding mix

    bn, by source

    LCR over the last 12 months

    month end, % · dashed = projected under the selected scenario

    Market risk

    Sample data · 250 trading days to 30 Sep 2026 · VaR 99%, 1 day, € m

    Desk

    VaR 99%, 1 day

    Stressed VaR

    2008–09 stress window

    Backtesting exceptions

    250 days · green 0–4, amber 5–9, red 10+

    VaR limit used

    VaR backtesting

    daily P&L against −VaR, € m · all desks

    Interest rate and credit spread sensitivity

    € k per +1 bp · blue = gain, orange = loss

    VaR against limits

    % of limit used · warning at 80%

    Regulatory reporting calendar

    Sample data · as of 6 Nov 2026

    Quarter

    Submissions due

    Submitted

    First-run validation pass rate

    Open data issues

    Submission timeline

    preparation window, progress and remittance date

    Validation pass rate

    rules passed on the first run, % · target 98%

    Open data issues by source

    count at quarter end

    Construction projects

    Sample data · week 41 of 2026 · € m

    Type

    Active projects

    Total budget

    Spent to date

    Behind schedule

    Cost and schedule performance

    CPI × SPI · bubble size = budget · click a project

    Progress: Riverside Offices

    % complete by week · dashed = forecast

    Cost by trade: Riverside Offices

    forecast final cost against budget, € m

    Swipe, drag or use the arrow keys to change dashboard

    Sample data built for these demos. All names and figures are illustrative.

    Software

    Products we design, build and run.

    Our software studio applies the same engineering discipline as our data work to everything it builds: a website builder, software that takes construction projects from drawing to permit and planning, and AI agents calibrated on your own data.

    Website builder

    Launch and edit your website without waiting for a developer.

    Choose sections, add your content and publish. Pages stay fast, accessible and on brand, and your team can change them whenever they need to.

    • Ready-made sections for every type of page
    • Forms, SEO settings and analytics built in
    • Fast pages on phones, tablets and desktops
    • Your own domain, with hosting and security handled

    Construction software

    From the first drawing to permit, budget and planning.

    Draw the building on its plot. The software checks feasibility against the zoning rules and your budget, calculates the bill of materials, drafts the permit application and the statement of work, and estimates how long construction will take. Change the drawing and everything updates.

    • DrawPlans and sections to scale, on the real plot.
    • FeasibilityZoning rules and budget checked as you draw.
    • Bill of materialsQuantities and costs taken from the drawing.
    • Permit draftApplication text and annexes, ready for your architect.
    • Statement of workWork packages with scope and quantities for contractors.
    • Construction timeA week-by-week plan and a completion date.
    Linden Court · new buildingExample project · plot 30 m × 24 m Feasible
    Floors 3

    Drag the corner or edge handles to resize. Arrow keys work on a focused handle.

    Example project with sample zoning rules, unit rates and productivity figures. Excludes VAT and fees.

    AI and machine learning

    Giving meaning to data, images and text.

    We build machine learning models and language-model agents that read your figures, documents, plans, maps and photos, and turn them into forecasts, standard documents, scores and checked answers. Each one is trained or calibrated on your own data, measured before it goes live and shows what it relies on.

    • Tables and figures
    • Text and reports
    • Images
    • Maps and parcels
    • Plans and schedules
    1 / 6

    Predictive models

    Sample model · 48,000 SME loans · neural network 5-7-5-1

    A neural network learns from past loans which businesses default within a year. Move the sliders to score a borrower, or train the network again and watch it learn.

    Borrower

    Pick a sample or move the sliders

    12
    2.4×
    1
    +6%
    80%

    Neural network

    Trained · 40 epochs · validation AUC 0.86

    Prediction

    Probability of default within 12 months

    Why this score

    Effect of each input against an average borrower, in points

    Learning curve

    Loss on training and validation loans, by epoch

    Work statements

    Sample quotes · large language model · standard library of work packages

    Contractors describe the same work in many ways, often in several languages. A language model maps each quote line to your standard work packages, normalises the units and lists what still needs confirming.

    Output language
    Standard lines8
    Units normalised4
    To confirm with the contractor3

    Quote lines, as received

    Three contractors · PDF, email and a site note

      Instructions to the model

      Standard statement of work

      Parcel analysis

      Sample district · 50 parcels · cadastre, zoning, flood and transit layers

      Each parcel is scored on what it lets you build, its distance to the station, its flood exposure and its price per m². Choose a profile or change the weights and the ranking updates.

      District map

      Colour = score for the selected profile · click a parcel

      Profile and weights

      What matters for this buyer

      Top 5 parcels

        Planning analysis

        Sample project · 9 work packages · 2,000 simulated schedules

        A model trained on your past projects estimates how long each task really takes, finds the critical path and simulates thousands of schedules to show how likely the deadline is. Delay a task or apply a suggestion to see the effect.

        Expected completion
        With 80% certainty
        Chance to meet the deadline

        Schedule and critical path

        Completion forecast

        What the model suggests

        Apply a suggestion to update the schedule and the forecast

          Contract review

          Sample subcontract · 8 articles · checked against your negotiation playbook

          A language model reads the contract article by article, compares it with your playbook and explains each risk with a proposed redline. Switch sides to see how the same clause reads for the other party.

          We are the

          Subcontract agreement · Linden Court

          Sample contract · extract

            Image agents

            Sample cases · calibrated on 400 decided cases

            An agent reads photos, maps and documents for one business decision and acts alone only when its calibrated confidence is high enough. Below that, a person decides.

            Agent benchClaim C-26-1187 · Wavre Acts alone

            Sources the agent may use

            Agent trace

              Calibration on 400 cases your team already decided

              90%
              Handled alone
              49%
              Right when alone
              96.4%
              Errors let through
              7

              Swipe, drag or use the arrow keys to change example

              Illustrative models, documents and figures built on sample data.

              Approach

              How a project runs

              The same four steps for every engagement, whether we automate a regulatory return or build a product.

              1. 01

                Map

                We map your sources, reports, controls and pain points, and agree what success looks like.

              2. 02

                Design

                We design the target data flow or product, including architecture, checks, screens and a plan with clear milestones.

              3. 03

                Build

                We build in short iterations, test against your real data and show working results along the way.

              4. 04

                Run

                We hand over documented, tested code, train your team and stay available for support and improvements.

              What you can count on

              Every number has a source

              Lineage links each reported figure back to the transactions behind it, so questions from auditors take minutes instead of days.

              Checks run before people do

              Validation and reconciliation run automatically, so your team's review time goes to judgement instead of ticking boxes.

              You own what we build

              Documented code and data models, handed over with training. No lock-in, and no black boxes in your reporting chain.

              Contact

              Tell us what you want to automate or build.

              Describe the report, process, dashboard or product you have in mind. We will reply with questions or a proposal for a first conversation.

              Email hello@celaconsulting.com

              Enter your name.

              Enter an email address, like name@company.com.

              Interested in

              Tell us briefly what you need.

              Prepares an email you can copy or open in your email app.