AI Feature Mapping Service — OpenLM Platform

🧠

Key Insight

OpenLM Platform already collects feature names from every SaaS customer's license servers. The Parser web tool processes thousands of license file uploads. This data is a goldmine sitting unused. Every parse event, every Broker checkout, every Products/Packages entry can feed an AI-powered mapping engine that gets smarter with every interaction — benefiting all customers.

1. Problem Statement

What license servers report

86445ACD_2025_0F

meba_r

nuke_i

Distrib_Computing_Toolbox

What customers understand

AutoCAD 2025

ANSYS Mechanical Enterprise

Nuke (Interactive)

MATLAB Parallel Computing Toolbox

The Gap

Today, each customer must manually map features to product names in the Products/Packages UI (AN-3022). This knowledge never transfers between customers. An AI service that learns from all customers' data can automate this mapping across the entire platform.

2. Approach: Supplement, Don't Replace

✅ What we build

• AI-powered feature → product mapping engine
• Kafka microservice in OpenLM Platform stack
• REST API for Parser web tool
• Cross-customer learning (anonymized)
• Vendor-specific rule engine
• LLM inference for unknowns

❌ What already exists (we don't rebuild)

• Parsers for 100+ license managers
• License collection (Brokers)
• Reporting (Spark + BI)
• Web UI (Products/Packages)
• Authentication / multi-tenancy
• Kafka infrastructure

Scope reduction: By plugging into OpenLM Platform, this is ~30% of a standalone product. We add the brain; the platform provides everything else.

3. Architecture in OpenLM Platform

Existing OpenLM Platform Components

EXISTING

Workstation Agents

Collect app data

EXISTING

Brokers

License server data

EXISTING

Parser Web Tool

Upload & parse files

▼ Kafka Topics ▼

NEW SERVICEAI Feature Mapping Service
Consumes
Feature events
PACKAGE structures
P&P corrections
Processes
Rule matching
AI inference
Confidence scoring
Produces
Product mappings
Confidence scores
Cost weights

▼ Enriched Events ▼

EXISTING

Enrichment Service

Merge + enhance data

EXISTING

Spark Aggregation

Reporting pipeline

EXISTING

BI Dashboard

Now shows product names!

Kafka Topic Integration

CONSUME broker.feature.checkout Feature checkout events from Brokers

CONSUME parser.result Parsed license file results

PRODUCE feature.mapping.enriched Enriched events with product mappings

CONSUME feature.mapping.feedback Corrections from Products/Packages UI

4. Mapping Pipeline

Four stages, cascading with early exit on high confidence (≥ 0.85):

①

Database Lookup

Have we seen this feature before? Check the mapping DB. Instant, zero cost.

Confidence: inherits from original mapping (0.5–1.0)

②

PACKAGE Analysis

Use PACKAGE → COMPONENTS relationships from the customer's license data. The PACKAGE name often IS the product name. High confidence because it's vendor-defined data.

Confidence: 0.90

③

Rule Engine

Vendor-specific pattern matching. Each vendor has naming conventions. Rules are versioned per release year.

Confidence: 0.85–0.89

④

AI Inference

LLM-based inference with few-shot context from known mappings of the same vendor. Uses local Qwen3-30B for cost efficiency.

Confidence: capped at 0.80 (requires human confirmation to go higher)

4.2 Vendor Catalog — 50+ Vendors Across 10 Industries

🏗️ CAD — Computer-Aided Design

Autodesk
AutoCAD, Inventor, Revit, Civil 3D
adskflex · ~500 features

Dassault (SW)
SolidWorks, SW Premium
sw_d · ~100 features

Siemens
NX, Solid Edge
ugslmd · ~300 features

PTC
Creo, Windchill
ptc_d · ~150 features

Bricsys
BricsCAD
bricsys (LM-X)

Nemetschek
Vectorworks, Allplan
vectorworks (LM-X)

⚙️ CAE — Simulation & Analysis

ANSYS
Mechanical, Fluent, HFSS, LS-DYNA
ansyslmd · ~200 features

Dassault (SIMULIA)
Abaqus, fe-safe, Tosca
ABAQUS / DSLS · ~150 features

MSC (Hexagon)
Nastran, Patran, Adams, Marc
msclmd · ~200 features

Altair
HyperWorks, OptiStruct, RADIOSS
altabortext · ~150 features

COMSOL
Multiphysics + modules
LMCOMSOL · ~50 features

ESI Group
PAM-CRASH, ProCAST
esilmd · ~80 features

BETA CAE
ANSA, META
BETA_LM · ~40 features

Dassault (CATIA)
CATIA, DELMIA, 3DEXPERIENCE
DSLS roles · ~200 features

🛢️ Oil & Gas / Geoscience

SLB (Schlumberger)
Petrel, ECLIPSE, Techlog, PIPESIM
slbsls · ~300 features

Halliburton
DecisionSpace, OpenWorks, SeisSpace
lgc_d · ~200 features

CGG (Viridien)
Hampson-Russell, GeoSoftware
cgg_d · ~100 features

AspenTech (Emerson)
Aspen HYSYS, Aspen Plus
aspentech · ~150 features

CMG
STARS, IMEX, GEM
cmg_d · ~30 features

IHS / S&P Global
Kingdom, Petra, AccuMap
ihs · ~80 features

Rock Flow Dynamics
tNavigator
rfd_d · ~20 features

dGB Earth Sciences
OpendTect
dgbes · ~20 features

🔌 EDA — Electronic Design Automation

Cadence
Virtuoso, Allegro, Spectre, Innovus
cdslmd · ~500+ features

Synopsys
DC, VCS, PrimeTime, IC Compiler
snpslmd · ~500+ features

Siemens EDA
Calibre, Questa, HyperLynx
mgcld · ~300 features

Altium
Altium Designer, Altium 365
altium (LM-X) · ~30

Keysight
ADS, PathWave, EMPro
aabortext · ~100 features

Zuken
CR-8000, E3.series
zuken · ~50 features

🏢 BIM + 🎬 DCC/VFX + 🔬 Scientific + 🧪 Process Engineering

Bentley
MicroStation, STAAD

Trimble
Tekla Structures

Foundry
Nuke, Mari, Katana

SideFX
Houdini FX

Chaos Group
V-Ray, Corona

MathWorks
MATLAB, Simulink

ESRI
ArcGIS Pro

AVEVA
E3D, PDMS

AspenTech
HYSYS, Aspen Plus

Intel
oneAPI, VTune

NVIDIA HPC
HPC SDK, PGI

ARM
Compiler, DS-5

Total Vendor Coverage 50+ vendors · 10 industries · ~5,000+ documented features

4.3 Rule Engine — Vendor Patterns

Autodesk (FlexNet, daemon: adskflex)

Format: {product_key}{code}_{year}_{ver}F · Source: Official Feature Code Lookup

86445ACD_2025_0F → AutoCAD 2025

87048INVNTOR_2025_0F → Inventor Pro 2025

86644RVT_2025_0F → Revit 2025

86760C3D_2025_0F → Civil 3D 2025

ANSYS (FlexNet, daemon: ansyslmd)

Abbreviated names, lowercase · Source: ANSYS License Management Guide

meba → Mechanical Enterprise

fluent → Fluent (CFD)

hfss → HFSS (EM simulation)

acfd_solvers → CFD Solvers

MathWorks (FlexNet, daemon: MLM)

MATLAB → MATLAB

Simulink → Simulink

Optimization_Toolbox → Optimization Toolbox

Signal_Toolbox → Signal Processing Toolbox

Foundry (RLM, ISV: foundry)

nuke_i → Nuke (Interactive)

nuke_r → Nuke (Render-only)

nukex_i → NukeX (Interactive)

mari → Mari

4.4 AI Inference Stage

For features no rule can resolve. Few-shot prompting with known examples from the same vendor.

# Prompt template
You are an expert in engineering software licensing.

Given feature {feature_name} from daemon {vendor_daemon}:

Known mappings for this vendor:
{known_examples}  # pulled from mapping DB

→ Returns: { product_name, confidence, reasoning }

Primary Model

Local Qwen3-30B (TRT-LLM)
Zero cost, fast, private

Fallback

Claude API / OpenAI
Complex/ambiguous cases

Safety: AI confidence capped at 0.80. Results shown as "Suggested: ..." until human confirms.

5. Cross-Customer Learning

The key advantage of being inside OpenLM Platform's SaaS platform:

Customer A

Broker reports meba_r → Rule engine maps to "ANSYS Mechanical Enterprise" (confidence: 0.88) → Stored

Customer B

Same feature meba_r → DB lookup → Already mapped! Zero computation. seen_count: 2

Customer C

Admin confirms in Products/Packages UI → Feedback event → confidence: 0.93 → ALL customers benefit

Privacy model: Only the mapping tuple crosses boundaries: (flexnet, ansyslmd, meba_r) → "ANSYS Mechanical Enterprise". No seat counts, hostIDs, usage data, or customer identifiers. This is factual product information, not customer data.

6. Confidence Scoring

0.95 – 1.00

Vendor-confirmed + 3 customer confirmations

0.90 – 0.94

PACKAGE-derived or rule + confirmed

0.85 – 0.89

Rule engine match (exact pattern)

0.50 – 0.84

AI inference (shown as "Suggested")

< 0.50

Unknown — not shown to users

Promotion: ≥0.85 auto-shown · 0.50–0.84 "Suggested: ..." with confirm button · <0.50 hidden

7. Integration Points

Parser Web Tool Enhancement

After parsing, one API call adds product names to the report:

Feature	Product NEW	Seats	Expiry
86445ACD_2025_0F	AutoCAD 2025 ✓ 92%	10	2026-12-31
87048INVNTOR_2025_0F	Inventor Pro 2025 ✓ 88%	5	2026-12-31
xyzCustom_Feature	? Unknown [Suggest]	2	2026-06-30

[✓ All Correct] [✗ Fix Mapping] [Export with Products]

Every "Correct" click = feedback event → improves mappings for all users.

Products/Packages UI Enhancement (AN-3022)

Before (manual)

Admin sees feature names → types product names one by one → knowledge stays with this customer only

After (AI-assisted)

Fields pre-populated with AI suggestions → admin confirms/corrects → corrections feed back to improve suggestions for ALL customers

8. REST API

GET

/api/v1/mapping/lookup?manager=flexnet&daemon=adskflex&feature=86445ACD_2025_0F

Single feature lookup → product mapping + confidence

POST

/api/v1/mapping/lookup-batch

Batch lookup for Parser web tool (all features in one call)

POST

/api/v1/mapping/feedback

Submit confirmation / correction from any UI

GET

/api/v1/mapping/stats

Coverage stats: total mappings, avg confidence, by vendor

GET

/api/v1/mapping/export?vendor=autodesk&format=csv

Export for Products/Packages bulk import

9. Data Seeding

Inside OpenLM Platform (Day 1)

SaaS Broker feature events10,000+ features

PACKAGE → COMPONENTS2,000+ bundles

Products/Packages entries500-1,000 confirmed

Parser web tool uploads5,000+ features

External Vendor Docs (by industry)

CAD / PLM

Autodesk (official lookup tool)~500

PTC, Siemens NX, Bentley~550

CAE / Simulation

ANSYS (licensing guide per release)~200

MSC/Hexagon, Altair, COMSOL, ESI, SIMULIA~630

Oil & Gas / Geoscience

SLB (Petrel, ECLIPSE, Techlog)~300

Halliburton, AspenTech, CGG, CMG~480

EDA

Cadence, Synopsys, Siemens EDA~1,300+

DCC / VFX + Scientific + Process

Foundry, SideFX, MathWorks, ESRI, AVEVA~540

Total vendor-documented~4,500+

Expected Day 1 coverage ~3,000 mappings

Expected Week 1 ~8,000 mappings

Expected Month 1 ~15,000 mappings

10. Technology Stack

Language: Python 3.12+

API: FastAPI (async)

Database: PostgreSQL 16

Queue: Kafka (OpenLM Platform native)

AI Primary: Qwen3-30B (local TRT-LLM)

AI Fallback: Claude API

Deploy: Kubernetes (Helm)

Footprint: 512MB RAM, 0.25 CPU

11. Development Phases

Phase 1 Core Service — Weeks 1-2

FastAPI scaffold · PostgreSQL schema · DB lookup + Rule engine (top 5 vendors) · REST API · Seed vendor docs · Docker + Helm

Phase 2 AI + Kafka — Weeks 2-4

AI inference stage · Kafka consumers (Broker events, Parser results, feedback) · PACKAGE analysis · Confidence scoring · Import existing P&P data

Phase 3 Integration — Weeks 4-6

Parser web tool integration · Products/Packages UI auto-suggestions · Cross-customer learning · Monitoring dashboard · More vendor rules

Phase 4 Scale — Weeks 6-8

Batch optimization · Auto-rule generation from feedback · Vendor partner APIs · Cost weight estimation · Production hardening

12. Success Metrics

Metric	Launch	Month 1	Month 3
Features mapped	3,000	15,000	30,000+
Avg confidence	0.70	0.80	0.85
Auto-accepted (≥0.85)	40%	60%	75%
Vendors covered	5	10	20
API response	<50ms	<30ms	<20ms

AI Feature Mapping Service — Design Document v2.0

For OpenLM Platform · LMP Consulting s.r.o. · 2026-03-17

Authored by Koda 🐾 with Council input