Migration Center

The Migration Center is the AI-powered tooling that converts Polkomtel's legacy ETL estate — Informatica PowerCenter, Alteryx, SSIS, and DataStage workflows — into native DataFlow AI YAML pipelines, with automated parsing, rule-based plus LLM-assisted conversion, per-object confidence scoring, and data-parity validation.

Who uses the Migration Center

The Migration Center is used primarily by the Data Engineer persona (Anna Kowalska) and the Platform Admin persona (Katarzyna Zielińska). Engineers drive day-to-day conversions; admins oversee the overall migration program.

For Polkomtel the scope is large — 500+ PowerCenter workflows plus 50–100 Alteryx workflows, totaling roughly 550–600 assets to migrate.

Module layout

The Migration Center mounts at /migration (redirecting to /migration/import; entry src/pages/MigrationCenter.tsx, layout src/pages/migration/MigrationLayout.tsx). A left sidebar lists the four screens; a breadcrumb trail sits at the top.

+------------------------------------------------------------------+
| Home > Migration Center > Import Wizard                           |
+------------------------------------------------------------------+
| Sidebar (240px)         | Main content area                      |
|  ARROW Migration Center | +-----------------------------------+  |
|   > Import Wizard        | |                                   |  |
|   > AI Conversion        | |   (Screen content renders here)   |  |
|   > Validation Suite     | |                                   |  |
|   > Progress Tracker     | |                                   |  |
+-------------------------+--+-----------------------------------+  |

The four screens:

Conversion pipeline — what happens under the hood

The migration engine runs a fixed lifecycle on every uploaded file: uploaded → parsing → parsed → converting → validating → validated → completed | completed_with_warnings | failed.

Conversion is deterministic where possible. PowerCenter transform rules and their target nodes:

The release gate: a job is marked completed only if overall confidence is at or above 0.85, no object scores below 0.80, no object needs manual review, and there are no validation issues — otherwise it is failed.

Screen 1 — Import Wizard

Route: /migration/import — entry src/pages/migration/ImportWizardPage.tsx.

A four-step wizard with a horizontal step indicator.

+------------------------------------------------------------------+
|  (1) Upload --- (2) AI Analysis --- (3) Report --- (4) Convert    |
+------------------------------------------------------------------+

Step 1 — Upload Files

A full-width drag-and-drop zone (dashed border, hover and drag-over states) with a Browse Files fallback. Each dropped file auto-detects its type — PowerCenter XML (orange badge), Alteryx Workflow (blue badge), or Unsupported (red badge) — and a quick parse reports the object count ("Detected: 7 mappings, 42 transformations"). Each file is shown as a card with its icon, name, size, type badge, an upload progress bar, and a Remove button. When more than one file is added, a batch indicator notes that all files will be analyzed together. The Analyze with AI button activates once at least one file is ready.

Step 2 — AI Analysis

A full-width overall progress bar plus a four-stage vertical checklist — Parsing Source Files → Analyzing Objects → Checking Compatibility → Generating Report — each stage card showing pending / in-progress (spinner) / completed (green check) state. A live, scrolling, terminal-styled Analysis Feed streams color-coded messages (info/success/warning/error) as the engine works through each workflow. The wizard auto-advances to Step 3 when analysis reaches 100%.

Step 3 — Compatibility Report

Four summary cards across the top — Total Objects, Auto-Convertible (count and %), Manual Required (count and %), and Estimated Effort (hours). Below them:

• Workflow Assessment table — one row per workflow with a complexity badge (Low / Medium / Medium-High / High / Very High), object count, auto-convert %, estimated effort hours, and source/target systems.
• Object Type Breakdown — a horizontal stacked bar chart, auto-convertible portion in indigo, manual portion in amber; object types under 50% auto-convert are flagged red.
• Risk Items panel — collapsible severity-coded cards (high/medium/low), each with a description, the affected workflow, and an expandable recommendation.

A Start AI Conversion button moves to Screen 2.

Behind the scenes

The frontend api/migration.ts posts the multipart upload to the migration-engine /upload endpoint, which runs all conversion stages inline. The engine's parsers use lxml for XML/YXMD and a proprietary text parser for DataStage; the max_upload_size_mb limit is 50 MB.

Screen 2 — AI Conversion Dashboard

Route: /migration/conversion — entry src/pages/migration/ConversionDashboardPage.tsx.

+------------------------------------------------------------------+
| [Total Objects 150] [Auto-Converted 127 (85%)] [Manual 23 (15%)] |
+------------------------------------------------------------------+
| Conversion Status by Object Type (table)                          |
+------------------------------------------------------------------+
| Converted Pipeline List (cards)   | Confidence Distribution chart |
+------------------------------------------------------------------+

Three summary cards lead the screen — Total Objects, Auto-Converted (with a green donut), Manual Required (with an amber donut) — plus average confidence.

The Conversion Status by Object Type table breaks results down per transform type (Source Qualifier, Expression, Lookup, Filter, Joiner, Custom Java, Router, Other): converted/total, conversion rate with an inline bar, average confidence, and a status of complete / partial / flagged. Flagged rows (such as Custom Java relying on proprietary Siebel CDMA libraries) get a red left-border, a red row tint, a warning icon, and a flag-reason line.

The Converted Pipeline List shows a card per converted pipeline — original → converted name (e.g. df_sap_biuro_sprzedazy_plk.yaml), a circular confidence badge (green ≥90, amber 70–89, red <70), object counts, source/target system pills, a status badge, and an Open in Design Studio link.

A Confidence Distribution histogram bins pipelines by confidence range (0-50%, 50-70%, 70-85%, 85-95%, 95-100%).

Alteryx workflows convert the same way PowerCenter does: the engine parses the .yxmd file into a common WorkflowAST, the rule engine maps each tool to a DataFlow node, and anything without a rule or below 0.60 confidence falls through to the LLM converter. Alteryx-specific constructs such as environment macros (GetEnvironmentVariable used for PROD/Test routing) are flagged for manual review rather than auto-converted, and surface as red-bordered rows in the Conversion Status table and as Risk Items in the Compatibility Report. A converted Alteryx pipeline shows a blue Alteryx source-type badge to distinguish it from the orange PowerCenter badge.

Behind the scenes

The conversion produces DataFlow AI YAML pipelines. The YamlGenerator classifies nodes into sources/transformations/targets, builds depends_on linkage from edges, and injects three default quality checks (row_count, null_percentage, duplicate) plus per-sink schema-validation checks and a low-confidence reconciliation check for any mapping under 0.80 confidence.

Screen 3 — Validation Suite

Route: /migration/validation — entry src/pages/migration/ValidationSuitePage.tsx.

The Validation Suite proves data parity between the legacy source and the converted DataFlow AI pipeline.

A Test Runner control bar at the top exposes Run All Tests and Re-run Failed, a status dot (idle / running / completed), and progress text. Four summary cards follow — Total Tests, Passed, Failed, Pass Rate.

The Pipeline Comparison Table lists each pipeline with its source and target systems, source vs target row counts, a row-count-match flag, a checksum-match flag, a column-diff count, execution time, and a pass/fail status. Failed rows are red-tinted with a red left-border and expand to reveal per-column failure detail — the column name, expected vs actual value, row index, and diff type (value_mismatch, null_mismatch, type_mismatch, or missing_row).

Behind the scenes

api/migration.ts calls the engine's /jobs/{id}/validate endpoint, which runs a six-check validation suite — YAML syntax, SQL safety (scanning for DROP TABLE/DATABASE, TRUNCATE, ALTER TABLE, EXEC, xp_cmdshell), transform coverage at or above 50%, confidence at or above 60%, source-and-sink node completeness, and mapping-issue checks.

Screen 4 — Migration Progress Tracker

Route: /migration/progress — entry src/pages/migration/ProgressTrackerPage.tsx.

The Progress Tracker gives a program-level view of the entire migration: the overall phases, conversion velocity, a timeline, and outstanding risks. It is the screen the Platform Admin uses to report migration status, with effort estimated by a person-hour model (0.25h base per mapping, 2.0h per manual review, 1.0h per LLM-assisted conversion, 0.5h testing, 4.0h integration).

Click-path — migrate a legacy Informatica workflow end-to-end

1. Open /migration/import.
2. Drag the PowerCenter export (e.g. wf_E112.XML) onto the drop zone. The card shows an orange PowerCenter XML badge and a detected object count.
3. Click Analyze with AI. The wizard moves to Step 2; watch the four-stage checklist and the live analysis feed stream parsing, classification, and compatibility messages.
4. When analysis completes, the wizard auto-advances to the Compatibility Report. Review the four summary cards, the per-workflow complexity badges, the object-type breakdown chart, and the Risk Items panel — expand high-severity items (such as Custom Java transformations needing a Python UDF rewrite) to read the recommendation.
5. Click Start AI Conversion — the wizard navigates to /migration/conversion.
6. On the AI Conversion Dashboard, inspect the Conversion Status by Object Type table. Flagged object types (red) need manual attention; partial and complete types are mostly automated.
7. In the Converted Pipeline List, click a pipeline card's Open in Design Studio to inspect or fix the generated YAML pipeline.
8. Go to /migration/validation and click Run All Tests to verify data parity.
9. Review the Pipeline Comparison Table — expand any failed row to read the per-column failure detail, fix the conversion in Design Studio, then Re-run Failed.
10. Track overall program status on /migration/progress.

Migration sub-page map

API reference

The migration-engine is a FastAPI service (port 8091) mounted under /api/v1/migration and /api/migration. It uses rule-based pattern matching for deterministic transforms and falls back to the Anthropic SDK for complex transformations. Converted YAML pipelines are consumed directly by Design Studio.