Machine Learning Research Intern (LLMs & Computer Vision)

Deep Learning / Machine Learning Research Intern (LLMs & Vision Language Models)

📍 Location: Bangalore (In-office)

💰 Stipend: ₹45,000/month

⏳ Duration

• 2026 Graduates: 6 Months
• 2027 Graduating Students: 10–12 Months

About HyperVerge

HyperVerge builds AI that decides whether a person or business is who they claim to be.

Trusted by 100+ enterprises including Jio, Vodafone, Airbus, SBI, Bajaj, Grab, Groww, and CRED, our AI systems have processed 2 Billion+ identity verifications across India, the US, Africa, and Southeast Asia.

We're building frontier AI for document intelligence, fraud detection, identity verification, and financial risk using Large Language Models (LLMs) and Vision Language Models (VLMs).

This is a research-first internship, where 50–90% of your work will involve reading papers, designing experiments, fine-tuning models, and pushing the boundaries of production AI.

What You'll Do

• Fine-tune open-weight LLMs and Vision Language Models using techniques such as SFT, PEFT/LoRA, and Preference Alignment.
• Build AI systems for document understanding, OCR enhancement, forgery detection, multimodal reasoning, and identity verification.
• Read, reproduce, and extend state-of-the-art research papers.
• Design evaluation benchmarks and improve model robustness against real-world and adversarial data.
• Build scalable training pipelines and optimize models for production.
• Own problems end-to-end—from research and experimentation to deployment.

What We're Looking For

Must Have

• Strong Python programming skills.
• Excellent understanding of Deep Learning fundamentals.
• Hands-on experience with PyTorch (preferred), TensorFlow, or JAX.
• Good understanding of Transformer architectures and modern neural networks.
• Experience building deep learning models for vision, multimodal, document, or image-related problems.
• Comfortable training, debugging, and evaluating models—not just consuming AI APIs.

Bonus

Experience with one or more of the following

• LLM or VLM fine-tuning
• PEFT / LoRA
• RLHF or Preference Alignment
• Hugging Face ecosystem
• LLaVA or similar multimodal architectures
• Distributed training
• CUDA optimization
• Research publications or open-source contributions

Show Us What You've Built

We're interested in builders.

Share anything that demonstrates your work, including

• Research papers
• Open-source contributions
• Fine-tuned models
• GitHub repositories
• Technical blogs
• Competition solutions
• Personal AI research projects

We care far more about what you've built than where you've studied.

This Role Isn't a Good Fit If...

Your experience is primarily

• Calling hosted LLM/VLM APIs without training or fine-tuning models.
• Limited to classical machine learning (Regression, Random Forests, XGBoost, etc.).
• Based entirely on tutorials, Kaggle notebooks, or guided coursework.
• Focused only on MLOps, backend infrastructure, or data engineering without model development.
• Built entirely through bootcamps, MOOCs, or virtual internships with no self-driven projects.

Eligibility

• 2026 Graduates available for a 6-month full-time internship.
• 2027 Graduating Students available for a 10–12 month full-time internship.
• Open to undergraduate and postgraduate students across all disciplines.
• No restrictions on degree, branch, CGPA, or backlogs. If you have relevant experience and can demonstrate your skills through our hiring process, we encourage you to apply.
• Ability to work from our Bangalore office for the duration of the internship.

What You'll Get

• Work on production AI systems serving 2 Billion+ identity verifications.
• Conduct applied research on state-of-the-art LLMs and Vision Language Models.
• Mentorship from engineers and researchers building production AI systems every day.
• High ownership from your very first week.
• Opportunity to solve challenging, real-world AI problems and see your work reach production.
• Exceptional performance during the internship may lead to a full-time opportunity at HyperVerge.