New opportunity

AI/ML Engineer

IQWorks · Mumbai, India

About this role

IQWorks seeks an AI/ML Engineer in Mumbai, India, to develop the models behind its privacy and data-classification platform. The role focuses on detecting personally identifiable information (PII), recognizing entities, classifying data and understanding documents across structured and unstructured sources. A key goal is enabling enterprises to process sensitive data within their own infrastructure through on-premise and edge deployment.

Responsibilities

Build, fine-tune and adapt NLP models for PII detection, named entity recognition and text classification; maintain the platform’s data-classification capabilities.
Create evaluation frameworks measuring precision, recall and accuracy across data types. Build training-data pipelines and labeling workflows to improve model quality.
Optimize models for small-footprint, low-latency, offline-capable on-premise and edge inference, as well as latency, throughput and cost efficiency in cloud production workloads.
Work with product and engineering teams to integrate AI capabilities. Research and prototype document understanding, contextual classification and anomaly detection approaches.

Requirements

At least 3 years of machine learning engineering or applied AI experience; strong Python proficiency and experience with modern ML frameworks.
Hands-on NLP experience in tasks such as named entity recognition, text classification, sequence labeling or document understanding, plus production model deployment in cloud and on-premise or edge environments.
Understanding of model training, evaluation, overfitting, data quality and the model lifecycle; familiarity with vector databases, embeddings and retrieval-augmented generation patterns.
Ability to use AI to accelerate research and delivery while exercising human oversight, solve problems independently and work in a fast-moving team.

Preferred

Experience in privacy or compliance-related ML, model quantization or distillation for edge deployment, OCR or document-processing pipelines, LLM fine-tuning, prompt engineering, agentic AI workflows, Go or TypeScript backend services, or shipping ML products end-to-end at a high-growth startup. The position is full-time.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

PythonMachine learningNatural language processingPII detectionNamed entity recognitionText classificationSequence labelingDocument understandingData classificationModel evaluationEdge deploymentOn-premise deploymentCloud deploymentModel inference optimizationTraining data pipelinesData labelingAnomaly detectionVector databasesEmbeddingsRetrieval-augmented generationModel quantizationModel distillationOCRLLM fine-tuningPrompt engineeringAgentic AIGoTypeScript

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

PII and classification models
The role builds and fine-tunes NLP models for PII detection, named entity recognition and text classification, and maintains data-classification capabilities across structured and unstructured sources.
Evaluation and training data
IQWorks calls for precision, recall and accuracy measurement across data types, plus training-data pipelines and labeling workflows to improve model quality.
Private on-premise and edge inference
A key focus is deploying small-footprint, low-latency, offline-capable models so sensitive data can be processed within enterprise infrastructure.
Cloud production efficiency
The role also optimizes cloud inference for latency, throughput and cost efficiency; the posting requires production deployment experience in both cloud and on-premise or edge environments.
Platform integration and prototyping
The engineer works with product and engineering teams to integrate AI capabilities, while prototyping document understanding, contextual classification and anomaly detection approaches.

Likely challenges to think through

Balancing missed detections and false alarms
Preparation hypothesis, not a confirmed IQWorks problem: PII detection across varied data may force precision–recall trade-offs. Prepare to explain how you would evaluate results by data type, inspect errors and use labeling feedback to guide improvements.
Preserving quality under edge constraints
Preparation hypothesis: reducing model footprint and latency for offline inference could weaken detection quality. Prepare a comparison of model-quality and resource metrics, with a clear approach to validating any optimization before deployment.
Reconciling cloud serving goals
Preparation hypothesis: improving throughput or lowering cost may increase latency in cloud workloads. Prepare to describe how you would benchmark these measures together and choose among competing serving configurations.
Moving prototypes into the platform
Preparation hypothesis: exploratory document-understanding work may compete with dependable integration and delivery. Prepare to explain how you would set evaluation criteria for a prototype, coordinate with product and engineering, and retain human oversight when using AI to accelerate the work.

Practice your answers

Technical practice

How would you approach PII detection as a sequence-labeling problem?

What to coverRole focus: Building and adapting NLP models for PII detection and named entity recognition. Preparation suggestion: Compare label design, context-dependent ambiguity, and ways to handle missed or incorrectly flagged entities.

How would you classify sensitive data across structured and unstructured sources?

What to coverRole focus: Maintaining data-classification capabilities for both source types. Preparation suggestion: Explain how fields, surrounding text, and document context might inform classification, and where one approach may fail.

How would you evaluate a PII detector across different data types?

What to coverRole focus: Designing evaluation frameworks for precision, recall, and accuracy across data types. Preparation suggestion: Discuss representative test sets, per-type results, and the trade-off between missed sensitive data and false positives.

How would you improve a training-data and labeling workflow?

What to coverRole focus: Building pipelines and labeling workflows to improve model quality. Preparation suggestion: Address label consistency, difficult examples, quality checks, and how evaluation findings would guide the next data collection or labeling cycle.

How would you diagnose overfitting during model development?

What to coverRole focus: Understanding training, evaluation, overfitting, data quality, and the model lifecycle. Preparation suggestion: Explain what you would compare across training and held-out data, then how you would distinguish model changes from data-quality fixes.

How would you make a model suitable for offline on-premise or edge inference?

What to coverRole focus: Small-footprint, low-latency inference that keeps sensitive data within enterprise infrastructure. Preparation suggestion: Discuss deployment constraints and how you would test accuracy against footprint and latency; quantization or distillation are listed as nice-to-have techniques.

How would you optimize inference in a cloud production workload?

What to coverRole focus: Improving cloud inference latency, throughput, and cost efficiency. Preparation suggestion: Describe how you would measure those factors together and evaluate whether an optimization harms prediction quality.

How would you prototype document understanding for varied documents?

What to coverRole focus: Researching document understanding; OCR and document-processing pipelines are nice to have. Preparation suggestion: Explain how you would assess extraction quality, preserve useful document context, and identify errors that affect downstream classification.

How would you evaluate contextual classification or anomaly detection?

What to coverRole focus: Prototyping both approaches. Preparation suggestion: Define what useful context or an anomaly would mean for a data-classification use case, then describe a baseline and evidence needed before adopting a prototype.

Where might embeddings, vector databases, or RAG fit in this work?

What to coverRole focus: Familiarity with embeddings, vector databases, and retrieval-augmented generation patterns. Preparation suggestion: Describe a plausible retrieval use case for document understanding, while weighing retrieval quality, latency, and sensitive-data handling rather than assuming the platform uses a particular design.

Behavioural practice

Tell me about collaborating to integrate an ML capability into a product.

What to coverRole focus: Working with product and engineering teams to integrate AI capabilities. Preparation suggestion: Use a real situation, your actions to resolve interface or quality trade-offs, and the result.

Describe a time you solved an ambiguous ML problem independently.

What to coverRole focus: Independent problem-solving in a fast-moving team. Preparation suggestion: Give the situation, how you scoped and tested options without waiting for complete direction, and the outcome.

When have you used AI to accelerate work while retaining human oversight?

What to coverRole focus: Using AI to speed research, experimentation, and delivery without compromising quality. Preparation suggestion: Share a real situation, what you delegated or verified yourself, and the result of those checks.

Tell me about a model-quality problem you investigated after deployment.

What to coverRole focus: Production deployment, evaluation, data quality, and the model lifecycle. Preparation suggestion: If you have a relevant example, explain the situation, your diagnostic and corrective actions, and the measured result; do not imply a specific incident at IQWorks.

Describe a time a prototype did not justify production adoption.

What to coverRole focus: Researching and prototyping new ML approaches. Preparation suggestion: Use a real situation to explain your evaluation criteria, how you communicated the trade-offs, and what happened next.

Sharpen your knowledge

Topics to prepare

PII detection and NLP task design
The role explicitly covers PII detection, named entity recognition, sequence labeling, and text classification. Suggested exercise: explain when span-level detection versus whole-text classification is appropriate, including how each would handle ambiguous sensitive data.
Evaluation and model quality
The posting calls for precision, recall, and accuracy across data types, plus understanding of overfitting and data quality. Suggested exercise: work through a hypothetical evaluation table and explain which errors matter most for PII detection and how you would investigate them.
Cloud versus on-premise and edge inference
Production deployment in cloud and on-premise or edge environments is required. Suggested exercise: compare trade-offs among model footprint, offline operation, latency, throughput, and cost for the same classifier in each setting.
Embeddings and retrieval-augmented generation
Familiarity with vector databases, embeddings, and RAG patterns is required. Suggested exercise: diagram a retrieval flow and identify where retrieval or generation could produce an incorrect classification. This is conceptual preparation, not a stated platform design.
Document and contextual classification
The role includes document understanding and prototyping contextual classification and anomaly detection. Suggested exercise: outline how document context could change a classification decision, and what evidence you would seek before adopting a prototype.

Skills to revise

Implement a Python NLP baseline
Strong Python proficiency, modern ML frameworks, and hands-on NLP experience are required. Suggested exercise: build a small PII span-labeling or text-classification baseline using practice data, then explain its training and inference steps.
Build a labeling and training-data workflow
The role includes training-data pipelines and labeling workflows. Suggested exercise: draft label definitions, check a small set of sample annotations for disagreement, and show how corrected labels would enter a repeatable training dataset.
Create a sliced evaluation report
The posting explicitly calls for an evaluation framework across data types. Suggested exercise: calculate precision, recall, and accuracy by data type and error category, then propose a targeted data or model change based on the results.
Test production inference trade-offs
The job requires cloud and on-premise or edge deployment, with different efficiency goals. Suggested exercise: design a repeatable benchmark for latency, throughput, footprint, and offline inference; explain what results would affect a deployment choice.
Prototype document processing
Document understanding is part of the role; OCR and document-processing pipelines are listed as nice to have. Suggested exercise: sketch a prototype that extracts text from a sample document and classifies sensitive content, noting where extraction errors could affect results. Treat OCR as optional preparation.
Review AI-assisted work
The posting expects AI-assisted research and delivery with human oversight. Suggested exercise: use AI to propose an experiment plan, then independently check its assumptions, evaluation criteria, and potential privacy-related errors before using it.

Your preparation checklist

Prepare an NLP project walkthrough
The role covers PII detection, NER, text classification, and document understanding. Prepare one relevant example, if available, showing the problem, Python and ML framework choices, model decisions, your contribution, and results.
Build an evaluation example
The posting calls for precision, recall, and accuracy across data types. Prepare a sample evaluation plan for structured and unstructured data, including error analysis, overfitting checks, and how findings would guide improvements.
Map a training-data workflow
Sketch how you would collect or generate examples, label them, check label quality, and feed corrections into retraining. Mark this as a proposed approach, not an IQWorks process.
Rehearse a deployment design
Outline how a PII model could run offline on-premise or at the edge while keeping sensitive data within enterprise infrastructure. Include footprint and latency trade-offs, then contrast cloud inference choices for throughput and cost.
Gather evidence for production and collaboration requirements
Identify examples you can substantiate of cloud and on-premise or edge deployment, model lifecycle work, independent problem-solving, and integration with product or engineering teams. If an example is unavailable, prepare an honest explanation of the gap.
Review adjacent techniques and responsible AI use
Prepare brief explanations of embeddings, vector databases, and RAG patterns, which are listed requirements. Be ready to explain where AI tools could speed experimentation and where you would retain human review; treat OCR and quantization as optional discussion topics.
Draft questions for the employer
Ask which data types and PII errors matter most, how model quality is measured, and what offline inference constraints the team faces. These are questions prompted by the role, not claims about IQWorks’ current systems.

Availability last checked: 5 Nov 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in India. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…