Ask for Python on MyInternships.in and 21,874 candidate profiles match. Ask for Machine Learning and the number is 3,281. Ask for Deep Learning and it is 975. Ask for NLP and it is 864.
Each step cuts the pool hard. Employers building AI teams often write a JD that asks for the last item on that list while budgeting for the first. Our numbers show why that rarely works.
The four steps side by side
These figures are drawn from 55,557 candidate profiles that list skills, counted on 1 Oct 2026. One caution up front: we count each label independently, so this is not a strict funnel in which every NLP profile is known to sit inside the Deep Learning group. It is a comparison of pool sizes.
| Skill as listed | Candidates | Share of 55,557 | Relative to the row above |
|---|---|---|---|
| Python | 21,874 | 39.4% | n/a |
| Machine Learning | 3,281 | 5.9% | 15.0% |
| Deep Learning | 975 | 1.8% | 29.7% |
| NLP | 864 | 1.6% | 88.6% |
The biggest drop is the first one. The Machine Learning count is 15.0% of the Python count. From there, Deep Learning is a little under a third of Machine Learning, and NLP is close behind Deep Learning.
Where the specialist profiles are
Bengaluru has the largest pool at every level: 344 Machine Learning profiles, 128 Deep Learning and 117 NLP. Pune follows on Machine Learning with 229, then Chennai with 208 and Hyderabad with 199.
The specialist labels thin out quickly outside those four cities. Delhi has 49 Deep Learning profiles and 45 for NLP. Mumbai has 39 and 44, making it one of the few places where NLP is listed more often than Deep Learning. Kolkata has 16 and 19.
In the smaller cities the counts fall to single or low double digits. Indore has seven Deep Learning profiles and nine for NLP; Lucknow has eight and seven. A locally bound specialist AI role in those cities is, on our platform, a search among very few people.
A common JD mistake
Many AI job descriptions list every term the team has heard of: machine learning, deep learning, NLP, computer vision and more. Each added term is treated by candidates, and by search filters, as a requirement.
Look at what that does to supply. A JD that only needs classical modelling, but mentions Deep Learning because it sounds current, has moved from a pool of 3,281 to one of 975 on our platform without any change in the real work.
Before publishing, ask the team lead which techniques the intern or fresher will use in the first few months. Put only those in the required list. Move the rest to a "you will get exposure to" line, which attracts ambitious candidates without excluding capable ones.
Machine Learning as a label is broad
The 3,281 figure deserves its own caution. Machine Learning on a fresher profile can mean a completed course with a standard project, a competition entry, or genuine applied work in an internship. The tag alone will not tell you which.
Useful questions to separate them:
- What was the baseline, and how much did your model improve on it?
- How did you split the data, and why that way?
- What would make this model fail once it left your laptop?
- Which feature mattered most, and how do you know?
Candidates who have only followed a tutorial tend to describe the algorithm. Candidates who have done the work describe the data and the mistakes.
Ask also to see the work. A notebook or repository with the candidate's own comments, including the dead ends, is far more convincing than a certificate. If everything in it ran perfectly the first time, it was probably not their own experiment.
Growing your own specialists
Given how small the specialist pools are, many teams do better hiring from the broader group and training upwards. NumPy is listed by 8,496 candidates and Pandas by 8,056. Those people already handle data in Python, which is the hard prerequisite for any modelling work.
A structured internship with a narrow, real problem is a sound way to turn a data-capable generalist into an NLP or deep-learning contributor. It also gives you a longer look at the person than any interview.
Be honest with candidates about this path. Say that the role starts with data preparation and evaluation and moves towards modelling as they show progress, so that nobody joins expecting to train large models in the first week.
What we are and are not saying
This is a tally of terms candidates chose for their MyInternships.in profiles. It is not an audit of capability and it should not be quoted as the size of India's AI workforce. People who do NLP work may simply write "Machine Learning", and the reverse. The shape of the narrowing is the reliable part.
Sizing an AI hire realistically
Match the JD to the true work, check the city-level pool before fixing a location, and decide whether you are hiring a specialist or developing one. Further supply data is on recruiter intelligence, and JD templates are in employer tools.
For early-stage teams, startup internships are a practical route to candidates who want hands-on AI work. To open a role now, hire AI and data interns on MyInternships.in.


