CONNECT DATACSV · TSV · XLSX · PARQUET
DROP DATA HERECHOOSE ONE LOCAL FILE
or click anywhere in this area to browse
Raw data stays on this device in Local Mode.
READY FOR INPUT
EVA DataLab: local data preparation, analytics and model evaluation demo
EVA / DATALABLOCAL WORKSPACE
LOCAL MODEDATA STAYS ON THIS DEVICE
PUBLIC INTERACTIVE BUILD · MAX 5 MB / 50,000 ROWS / 50 COLUMNSBLUEPRINT UI / WORKING COREAvailable features remain interactive · planned V1 stages are marked Coming soon
DATA SOURCE
PROJECT WORKSPACE.
DATASETLOADED CSVLocal CSV upload
01 / CONNECTLOCAL MODE
BRING IN
YOUR DATA.
Choose one local data file (CSV, TSV, XLS/XLSX or Parquet). EVA reads and analyses the selected file in this browser, subject to demo limits. Excel and Parquet parsers are loaded from third-party CDNs; avoid confidential data in this public demo.
IMPORT NEEDS ATTENTIONThis file is outside the public build scope.
Discuss a larger solution ↗CURRENT SOURCELOCAL MODE
CONNECTED
Dataset loaded in this browser.
The source is immutable. Later stages work from a separate working copy.
FORMAT—
FILE SIZE—
SHEET / TABLE—
LOADED—
Import details + project state
IMPORT CAPABILITIES01 / CONNECT
CSVComma-delimited textREADY
TSVTab-delimited textREADY
XLSXWorkbook + sheet selectionREADY
PARQUETColumnar analytical fileREADY
SAVED PROJECTOpen a reusable EVA projectDELIVER DEPENDENCY
PROJECT STATE
—CREATED ON IMPORT
The source is preserved as the immutable reference for this browser session.
IMPORT CONFIRMATIONPREVIEW ONLY · SOURCE UNCHANGED
ROWS—
COLUMNS—
STATUSVALIDATED
The first rows are shown only to confirm that the file was read correctly.
READING FILEDETECTING COLUMNSPROFILING QUALITYBUILDING PREVIEW
IMPORT NEEDS ATTENTION
02 / PROFILECOMPLETE CORE
UNDERSTAND WHAT
YOU HAVE.
Profile explains the dataset before anything is changed. EVA uses deterministic rules to identify structure, quality, likely column roles and the statistics that matter for each data type.
PROFILE SUMMARY
Dataset profile ready.
Choose a column below to inspect its role, type, completeness and distribution.
COLUMN MAPSELECT A COLUMN TO INSPECT
FIELDTYPEROLEQUALITY
COLUMN INSPECTORWAITING
SELECT A FIELDInspect one column at a time.
EVA will show the detected type, semantic role, missing values, unique values and type-specific statistics.
—
——
—
DISTRIBUTIONRULE-BASED
DATA PREVIEWFIRST 8 ROWS · SOURCE UNCHANGED
How EVA profiled this
HOW EVA PROFILED THISLOCAL · DETERMINISTIC
TypesValues are checked locally for numeric, date and text patterns.
RolesIdentifiers, categories, measures, dates, constants and possible outcomes are flagged with rules, not a cloud LLM.
Source safetyProfiling reads the source but does not modify it.
03 / PREPAREGUIDED CLEANING
FIX WHAT NEEDS ATTENTION.
ONE COLUMN AT A TIME.
EVA turns the Profile findings into a guided cleaning queue. Choose how each issue should be handled, preview the exact impact, then apply it as a reversible recipe step.
Cell-level issues detected in the current working copy.
— cells total— dataset-level
REVIEW QUEUEANALYSING…
EVA is checking the working copy for column-level and dataset-level issues.
—columns—issues0ignored
COLUMNS TO REVIEWWORKING COPY
Choose a column or dataset issue.
EVA only shows issues it can explain. Nothing is changed until you preview and apply a method.
TRANSFORM DATAOPTIONAL TOOLS
Use a transformation when you already know what you want to do.
These tools are separate from the guided quality queue so sorting, filtering and reshaping do not look like errors.
ROWS
Filter · Sort · Sample · Remove duplicates · Remove incomplete rows
TEXT
Trim · Change case · Find/replace · Remove symbols · Fill missing
NUMERIC
Fill missing · Cap values · Remove outliers · Round
STRUCTURE
Convert type · Rename · Drop column
FIX SELECTED ISSUECHOOSE FROM THE LEFT
SELECT AN ITEMChoose a column to review.
Select an issue from the left. EVA will show only the relevant cleaning methods.
REVIEW
—
—
HOW SHOULD EVA HANDLE THIS?
Choose a cleaning method.EVA will explain what changes before anything is applied.
CURRENT VALUESCHOOSE AN ISSUE
CURRENT——
→PREVIEW—Choose a method first
BEFORE · AFFECTED VALUES
AFTER · AFFECTED VALUES
More quality diagnostics
ISSUES DETECTEDWORKING COPY
MISSING VALUESBY COLUMN
COLUMNS TO REVIEWIDENTIFIERS · CONSTANTS · OUTLIERS
RECIPE BUILDERWORKING · LOCAL
NON-DESTRUCTIVE EXECUTIONSource data stays unchanged. EVA rebuilds the working copy by replaying enabled steps from the original source.
04 / EXPLOREASK THE DATA
ASK THE DATA
WHAT MATTERS.
AUTO-DETECTED—Choose a target.
TARGET VIEWTASK VIEW
EVA / SIGNALEXECUTIVE READ
WAITINGChoose a target.
EVA will summarise the strongest signal and model readiness.
More analysis + methodology
FEATURE VIEWCOMPARISON
MODEL READINESSPRE-MODEL CHECKS
AUTO VISUALSELECTED FROM FIELD TYPES
FEATURE ASSOCIATIONSTARGET-AWARE
HOW TO READ ITINTERPRETATION
QUALITY BOUNDARYCAVEATS
METHOD NOTEEDA ≠ MODEL EVIDENCE
05 / MODELPREDICT AN OUTCOME.
PREDICT AN OUTCOME.
TEST THE EVIDENCE.
DETECTED TASK—Choose what you want to predict.
RECOMMENDED WORKFLOWChoose a target.
EVA will suggest a starter model and explain why.
EVALUATION VIEWHELD-OUT TEST SET
EVA / MODEL READPLAIN ENGLISH
NOT TRAINED
Train a model.
This card will explain performance, the baseline and the most important weakness.
Validation + technical details
MODEL COMPARISONSAME HELD-OUT SPLIT
Build a model to compare the available algorithms.
5-FOLD VALIDATIONSTABILITY CHECK
Build a model to estimate fold-to-fold stability.
METHODHOW THE TEST WAS RUN
EVA uses an 80/20 held-out split and task-appropriate metrics.
NEXT CHECKSBEFORE USING A MODEL
Train a model to receive validation guidance.
06 / PREDICTCOMING SOON
TURN THE MODEL
INTO A TOOL.
This stage is intentionally visible now so the product journey is clear. It will reuse the exact preprocessing and selected model from the earlier steps.
WHAT-IF PREDICTIONONE SCENARIO
RESULTPredicted value / class
Classification will show class + probability where supported. Regression will show a numeric result with context.
BATCH PREDICTIONNEW RECORDS
＋DROP NEW DATASchema validation before scoring
01Check required fieldsPLANNED
02Apply training preprocessingPLANNED
03Generate predictionsPLANNED
04Export resultsPLANNED
DEPENDENCYPrediction will only unlock after a model and reusable preprocessing pipeline exist.
This prevents training and prediction from silently using different transformations.
07 / DASHBOARDCOMING SOON
MAKE RESULTS
LIVE + EXPLAINABLE.
The V1 dashboard will combine prepared business data and model output in one no-code, filterable view. This is a UI preview only.
PREVIEW
AVAILABLE NOW
Decision brief + cleaned CSV export are working. Project packages, recipe files and additional export formats are represented below as planned V1 capabilities.
08 / DELIVERSAVE. SHARE.
SAVE. SHARE.
RUN IT AGAIN.
INSIGHT CONTEXT—Insights follow the selected target and model.
EXECUTIVE SUMMARYAUTO-GENERATED
WAITING
Select a target.
The summary will connect data quality, signal strength and model evidence.
DECISION CONTEXTWHAT THE ANALYSIS SUPPORTS
MODEL INTERPRETATIONEVIDENCE
RECOMMENDED ACTIONSNEXT BEST STEPS
RISKS + CAVEATSREAD BEFORE ACTING
AI + HUMAN / NEXT LEVELCUSTOM PROJECT
Need this on larger or production data?
The public DataLab is deliberately small. EVALUATION can build database connections, automated pipelines, secure server-side analytics and production ML workflows.
Start a project ↗FULL BUSINESS DEMOWant to see a complete management workspace?
Open EVA Workspace ↗Open the KISSA EVA Workspace to see dashboards, forecasting and management actions around a full business scenario.
EXPERT MODE / ADVANCED TOOLSSQL
SQL
EXPLORER.
QUERYLIGHTWEIGHT DEMO SQL
RESULTWAITING