NLP Classification in 2018 vs LLMs Today: The Interface Changed, the Engineering Didn't

In 2018 I ranked 1st in an IBM Watson AI Lab competition building an NLP classification workflow. Today a first prototype is much faster — but the engineering discipline behind quality is exactly the same.

NLPClassificationLLMWatson
NLP classification workflow: 2018 IBM Watson engineering versus modern LLM engineering

TL;DR: I do not think of this as "old AI versus new AI." It is the same engineering principles expressed through better tools. If your LLM output is unreliable, fix the instructions, the examples, and the contract — not the model.

The Problem

In 2018, I ranked 1st in an IBM Watson AI Lab competition. The task was to build an NLP classification workflow — and at the time, the path looked like this:

  • Define intent boundaries
  • Prepare and label training examples
  • Configure the Watson services
  • Build routing logic
  • Test edge cases
  • Iterate across multiple components

A working prototype took time because the system had many moving parts.

The Approach: Today

For a similar classification problem today, a first prototype can often start much faster:

  • Define the task clearly in a system prompt
  • Specify a structured output schema
  • Call a modern LLM
  • Validate the response in the application

The speed difference is real. But the more interesting lesson is what did not change.

What Did Not Change

In 2018, quality depended on

Clear intent boundaries · useful examples · explicit routing logic · reliable contracts between components

Today, quality still depends on

Clear instructions · useful context and examples · output schemas · well-defined workflow boundaries

The interface changed. The engineering discipline did not. A vague task definition still creates unreliable outputs. A poorly specified contract still breaks downstream systems. And a demo that works once is still not the same as a system you can operate.

Key Takeaway

Design insight: I do not think of this as "old AI versus new AI." It is the same engineering principles expressed through better tools. If your LLM output is unreliable, fix the instructions, the examples, and the contract — not the model.

Comments

FAQ

What is the key takeaway from "NLP Classification in 2018 vs LLMs Today: The Interface Changed, the Engineering Didn't"?

I do not think of this as "old AI versus new AI." It is the same engineering principles expressed through better tools. If your LLM output is unreliable, fix the instructions, the examples, and the contract — not the model.

Who wrote this and what is it about?

This was written by Mahmoud Trigui, Senior Data Scientist. In 2018 I ranked 1st in an IBM Watson AI Lab NLP classification competition. The tools have changed, but the engineering discipline that made it work hasn't — clear instructions, useful examples, and well-defined contracts still decide quality.

Machine LearningFeature EngineeringMLForecastTime Series DecompositionForecastingLightGBMXGBoostCatboostClusteringSegmentationNLPLLMsWeb AppR MarkdownSQLOracle DBSAS-GuideSAS E-MinerDataikuBigQueryGCPPythonRCRISP-DMHypothesis TestingANOVAData AnalyticsDimensionality ReductionRecommendation SystemNetwork AnalysisGeospace AnalysisSpatial DataEmbeddingSampling TechniquesDecision RulesData StorytellingCVMChurnFraud DetectionSentiment AnalysisTopic ModelingIBM WatsonPowerBILooker StudioVBAStatistical LearningEnsemble ModelingStackingCross-ValidationProfilingABT ConstructionPlumberTidyverseShinyProphetDeep LearningScikit-LearnJSONSAS ProgrammingGitVS CodeCSS StylingAutomated ReportingOutlier DetectionTemporal ClusteringStartup SurvivalPre-Valuation ModelingK-MeansDecision TreesData SciencePredictive ModelingSVMLDAText ClassificationWeight PredictionPattern RecognitionReal-Time DetectionCommunity DetectionPipeline AutomationData Quality ChecksData ReliabilitySpecification MappingBusiness StrategyMarketing CampaignsTry & Buy FrameworksKPI DashboardsNetwork QualitySales AnalyticsMentoringStatistics LecturerRemote WorkHybrid WorkConsultingContractFull-TimeFreelanceSofrecomOrange GroupTunisia TelecomKiota IntelligenceVC AnalyticsSeries A PredictionProduction MLApplied AIPrompt EngineeringBusiness ForecastingDecision SystemsGraph AnalyticsHousehold DetectionMulti-SIM DetectionFTTH ForecastingPydanticGPT-4OpenAI APIZindiCell Tower AnalysisBehavioral SignalsObservation Unit DesignWard ClusteringTarget Encoding Machine LearningFeature EngineeringMLForecastTime Series DecompositionForecastingLightGBMXGBoostCatboostClusteringSegmentationNLPLLMsWeb AppR MarkdownSQLOracle DBSAS-GuideSAS E-MinerDataikuBigQueryGCPPythonRCRISP-DMHypothesis TestingANOVAData AnalyticsDimensionality ReductionRecommendation SystemNetwork AnalysisGeospace AnalysisSpatial DataEmbeddingSampling TechniquesDecision RulesData StorytellingCVMChurnFraud DetectionSentiment AnalysisTopic ModelingIBM WatsonPowerBILooker StudioVBAStatistical LearningEnsemble ModelingStackingCross-ValidationProfilingABT ConstructionPlumberTidyverseShinyProphetDeep LearningScikit-LearnJSONSAS ProgrammingGitVS CodeCSS StylingAutomated ReportingOutlier DetectionTemporal ClusteringStartup SurvivalPre-Valuation ModelingK-MeansDecision TreesData SciencePredictive ModelingSVMLDAText ClassificationWeight PredictionPattern RecognitionReal-Time DetectionCommunity DetectionPipeline AutomationData Quality ChecksData ReliabilitySpecification MappingBusiness StrategyMarketing CampaignsTry & Buy FrameworksKPI DashboardsNetwork QualitySales AnalyticsMentoringStatistics LecturerRemote WorkHybrid WorkConsultingContractFull-TimeFreelanceSofrecomOrange GroupTunisia TelecomKiota IntelligenceVC AnalyticsSeries A PredictionProduction MLApplied AIPrompt EngineeringBusiness ForecastingDecision SystemsGraph AnalyticsHousehold DetectionMulti-SIM DetectionFTTH ForecastingPydanticGPT-4OpenAI APIZindiCell Tower AnalysisBehavioral SignalsObservation Unit DesignWard ClusteringTarget Encoding