How AI is closing India's admission counselor gap.

I want to write down a number I find difficult to forget. India has, depending on the year you measure, between 90 and 110 million students at any point in the senior secondary or undergraduate decision band. Against that, the All India Council for Career Guidance and the various private counselor-certification programmes together produce somewhere between 40,000 and 50,000 trained admission counselors. That's not a typo. The ratio is 2,000 students per counselor.

A trained counselor working at the upper edge of their craft — taking time to read each student's profile, weigh their constraints, sit with the family for a real conversation — can serve maybe 100 students a year. To serve all 100 million school-leavers we would need 1.2 million counselors. India does not have that many, and is not going to have that many in our lifetime.

The shape of the gap

The gap is not evenly distributed. In tier-1 metros, an upper-middle-class family can hire a private counselor for ₹15,000 – ₹40,000 for the JEE / NEET / CUET cycle. In tier-2 cities, the price drops but the quality often drops with it. In rural districts and in low-income urban households, the counselor option simply doesn't exist.

What fills the gap today, mostly, is family. Anyone who has been to college ends up as the household admissions counselor for cousins and neighbours. That worked when "college" meant five possible institutions and three possible streams. It does not work when the country's official college and scholarship catalog runs to 1,200+ UGC-recognised institutions, 8,400+ active scholarships, and stream choices are bifurcating into specialisations like Mech-AI and Civil-AI that didn't exist when the family elder graduated.

Why large language models change the ratio

The economics of human counseling are stuck. The economics of LLM counseling are not. A single GPU running an Anthropic Claude inference can simultaneously hold a counseling conversation with hundreds of students at a per-session cost that is fractions of a rupee. The hard problem is not capacity. The hard problem is quality — getting the agent to actually understand a student's situation, not give boilerplate advice, and not hallucinate.

What we've found running NIGOC in our closed beta is that the quality threshold is reachable, but you don't get there by wrapping a chatbot around a search engine. You get there by:

• Building specialised agents per stage of the journey. One agent recommends colleges. A different agent finds scholarships. A third predicts career trajectories. A fourth drafts and critiques SOPs. They share a knowledge base but they specialise. (We explain the five at our platform page.)
• Refusing to use English as default. Most counseling that matters happens in a language the family is comfortable in, in the room. We launch features in 22 Indian languages because if we don't, the agent serves the same students who already had access to counselors.
• Indexing what's actually open and active, nightly. Indian scholarship eligibility windows move week to week. The agent is only useful if it knows what's open today, not what was open 18 months ago.
• Being transparent about limits. No agent should pretend it's a human. We tell every student when the agent's confidence is low and recommend they consult a human counselor or institutional admissions office.

What we're seeing in beta

In 2,500 students through the closed beta, the outcomes data is encouraging:

• 94% of post-counseling survey responses rated recommendations "useful" or "very useful."
• Students found 2.8× more scholarships they were eligible for than students who searched on their own (control group n=110).
• Of students who came in saying "I will pick CS" and went through a Career Pathway session, ~22% changed their primary choice — sometimes to a specialised branch like Mech-AI, sometimes out of engineering entirely.

That last number is what gets me. It is the number that says the agent is doing what we wanted — surfacing alternatives a 17-year-old wouldn't have considered, with data instead of opinion.

What this is not

NIGOC is not a replacement for the kind of years-long mentorship a good human counselor can offer. We do not pretend to be. The Indian admission counseling problem has a long tail of high-touch needs — applying to top-10 US undergrad programmes, niche conservatory auditions, lateral entry across diploma-to-degree paths — where a human with specific knowledge and a network beats anything an LLM can do today.

What we are saying is that 90% of the volume of Indian admission counseling demand — the basic "which college, which stream, which scholarship" question, asked in the student's own language, with up-to-date data — is suddenly solvable. That was not true three years ago. It is true now.

The next decade of EdTech in India is going to be defined by who builds that solution well, who builds it badly, and who hits the regulatory and language barriers first. We are obsessed with the first option.