AI ML Intern

AI/ML Engineer Intern

Company: Ortus Innovation Pvt. Ltd.

Product: DXINT — DeepShadow Augmented Intelligence (Sugreev Engine)

Experience: 0–1 years | Freshers welcome

Location: Remote, India

Duration: 3–6 months

Type: Full-time internship

Schedule: 6 days per week

Stipend: To be disclosed

About Ortus Innovation

Ortus Innovation is building DXINT (DeepShadow Augmented Intelligence), an indigenous, AI-powered intelligence platform designed to transform large volumes of public-source information into structured, explainable, and actionable insights.

DXINT's Sugreev engine processes information across news, research publications, public digital sources, documents, and other media formats. Its AI/ML pipeline supports multilingual information extraction, entity and location identification, sentiment and narrative analysis, semantic clustering, pattern detection, source credibility assessment, and intelligence prioritisation.

We are building a modular, production-oriented platform where machine learning must work reliably on real-world, noisy, multilingual, and continuously changing data.

Role Overview

We are looking for a highly curious, technically strong, and execution-oriented AI/ML Engineer Intern to help build and improve the machine learning pipeline powering DXINT.

You will work on practical ML engineering problems involving NLP, multilingual text processing, embeddings, classification, semantic clustering, information extraction, model evaluation, and explainable scoring systems.

This is not a research-only or prompt-engineering role. You will be expected to implement working components, evaluate their performance, integrate them into pipelines, and contribute maintainable code to a real product.

You will work closely with the founding and engineering team, with opportunities to take ownership of individual ML components from experimentation through integration.

Key Responsibilities

1. NLP and Multilingual Intelligence Processing

• Develop and evaluate NLP pipelines for processing noisy, unstructured, and multilingual text.
• Implement text normalisation, language identification, translation integration, tokenisation, and preprocessing workflows.
• Experiment with Named Entity Recognition (NER), entity resolution, keyword and phrase extraction, and location extraction.
• Work with transformer models and libraries such as Hugging Face Transformers and SentenceTransformers.

2. Embeddings, Semantic Search and Clustering

• Generate semantic embeddings for documents, events, entities, and text segments.
• Implement semantic similarity, duplicate detection, and near-duplicate identification.
• Develop clustering pipelines using techniques such as DBSCAN, HDBSCAN, or other suitable algorithms.
• Group related information into meaningful themes, narratives, and event clusters while minimising false associations.
• Evaluate embedding and clustering quality on representative datasets.

3. Classification, Sentiment and Narrative Analysis

• Build and evaluate text classification models for relevance filtering and domain-specific categorisation.
• Develop contextual sentiment analysis and stance or narrative classification pipelines.
• Experiment with multi-label classification for configurable intelligence categories.
• Design approaches for distinguishing relevant signals from noise, misleading content, and irrelevant mentions.
• Evaluate traditional ML, transformer-based models, and LLM-assisted approaches where appropriate.

4. Entity, Location and Pattern Analysis

• Build extraction pipelines for people, organisations, locations, facilities, events, and other domain-relevant entities.
• Work on entity normalisation, alias matching, geolocation enrichment, and structured output generation.
• Explore temporal patterns, co-occurrence analysis, graph-based relationships, and anomaly detection.
• Help identify recurring patterns and emerging changes across collections of documents without treating unrelated records as correlated evidence.

5. Source Reliability, Confidence and Explainability

• Help implement source-quality signals, confidence estimation, provenance tracking, and evidence-linked model outputs.
• Develop transparent scoring components that distinguish model confidence, source reliability, relevance, and evidence strength.
• Ensure outputs can be traced back to their supporting records and transformations.
• Support human-in-the-loop review, auditability, bias checks, and explainability.

6. ML Pipeline Engineering and Integration

• Build modular, testable ML components using Python.
• Integrate models into APIs or asynchronous processing pipelines using FastAPI and related technologies.
• Work with PostgreSQL, PostGIS, vector databases, and object storage where required.
• Support batch inference, embedding generation, model versioning, logging, and performance monitoring.
• Profile latency, memory consumption, throughput, and inference costs.
• Write unit tests, evaluation scripts, reproducible experiments, and clear technical documentation.

7. Experimentation and Model Evaluation

• Establish meaningful baselines before introducing complex models.
• Prepare datasets, label samples where required, and analyse failure cases.
• Measure precision, recall, F1-score, false-positive rates, and latency according to the task.
• Compare models systematically and document trade-offs.
• Investigate failures involving ambiguous text, multilingual aliases, duplicate content, noisy inputs, and distribution shifts.

Technical Stack

Core language

• Python

Machine Learning and NLP

• PyTorch
• Hugging Face Transformers
• SentenceTransformers
• Scikit-learn
• Embeddings, semantic similarity, classification, clustering, and NER

Data Processing and Integration

• Pandas or Polars
• FastAPI
• PostgreSQL and PostGIS
• Vector databases such as Milvus
• Git and GitHub

Additional exposure

• DBSCAN/HDBSCAN
• LLMs and retrieval-augmented generation (RAG)
• MLflow or equivalent experiment tracking
• Docker and AWS
• OCR and speech-to-text pipelines
• Neo4j or other graph technologies

You are not expected to know every technology on day one. Strong Python and ML fundamentals, the ability to learn independently, and demonstrable implementation skills matter most.

Required Qualifications

• 0–1 years of experience in AI/ML, data science, NLP, or related engineering; freshers are welcome.
• Strong Python programming fundamentals, including data structures, functions, modular code, and debugging.
• Understanding of core ML concepts: training, validation, overfitting, evaluation metrics, and model generalisation.
• Familiarity with at least one ML framework, preferably PyTorch or Scikit-learn.
• Basic understanding of NLP, embeddings, classification, or information extraction.
• Experience with Git/GitHub and the ability to work with an existing codebase.
• At least one meaningful academic, personal, open-source, or internship project demonstrating practical technical execution.

Preferred Qualifications

• Hands-on work with transformer models, SentenceTransformers, or Hugging Face.
• Experience with semantic search, vector databases, clustering, or entity extraction.
• Exposure to multilingual NLP, translation, NER, sentiment analysis, or document processing.
• Familiarity with SQL, REST APIs, FastAPI, and data pipelines.
• Experience evaluating models on real datasets rather than relying only on demo outputs.
• Interest in intelligence analytics, information retrieval, graph analytics, or large-scale unstructured data processing.

The Mindset We Value

Ownership: You take a problem from investigation to implementation, testing, and documentation.

Curiosity: You want to understand why a model works, where it fails, and how to improve it.

Scientific thinking: You form hypotheses, establish baselines, measure outcomes, and avoid unsupported claims.

Resourcefulness: You read documentation, debug independently, and work through unfamiliar technical problems.

Diligence: You care about data quality, edge cases, reproducibility, and false positives.

Integrity: You understand that intelligence systems require traceable evidence, responsible data handling, and human oversight.

What You Will Work Towards

During the internship, your work may contribute to measurable deliverables such as:

• A tested multilingual text-processing or entity-extraction pipeline.
• An embedding-based semantic search or duplicate-detection component.
• A classification or clustering model evaluated against a defined baseline.
• A reproducible model evaluation framework with documented error analysis.
• An ML component integrated into a working application or data-processing pipeline.
• Technical documentation and performance measurements for the component you own.

Specific deliverables will depend on your initial skill assessment and the product roadmap.

What You Will Gain

• Hands-on experience building ML components for a real, evolving product.
• Exposure to end-to-end ML engineering: data preparation, experimentation, evaluation, deployment, and monitoring.
• Direct collaboration with the founding and engineering team.
• Opportunities to take ownership of production-oriented ML modules.
• A certificate of completion and a recommendation letter reflecting your demonstrated contributions, subject to successful completion.
• High-performing interns may be considered for a full-time engineering role, subject to performance and business requirements.

Work Details

• Location: Remote, India.
• Duration: 3–6 months.
• Schedule: Six days per week.
• Start date: Immediate / rolling.
• Stipend: To be disclosed.

The expected workload, stipend arrangements, and internship terms will be communicated clearly during the selection process.

Application Process

Please share:

1. Your CV.
2. Your GitHub profile, if available.
3. Links to relevant ML/NLP projects, notebooks, demos, or research work.
4. A short explanation of one technical problem you solved and how you evaluated your solution.

Shortlisted candidates may be asked to complete a practical ML assignment and explain their implementation, evaluation methodology, design choices, and limitations.

We value demonstrated ability, independent thinking, and execution over credentials alone.

If you enjoy turning messy data into reliable, measurable machine learning systems, we would like to hear from you.