Vahan.ai uses Nvidia AI to speed up blue-collar hiring
Vahan.ai has fine-tuned Nvidia’s Nemotron 3 Nano to build a faster, cheaper and more specialised voice-based recruiter for blue-collar workers.
Hiring at scale is difficult. For blue-collar jobs, the challenge is even bigger when recruiters need to reach thousands of candidates across cities, languages and skill levels.
Indian startup Vahan.ai is using AI to tackle this through its voice-based recruiter. The company has worked with Nvidia’s technical team to fine-tune Nemotron 3 Nano, a smaller AI model designed to handle recruitment conversations more efficiently.
A voice recruiter built for India’s workforce
Vahan.ai’s AI recruiter speaks to candidates over phone calls, helping them discover suitable jobs, apply for roles and schedule interviews. This can be particularly useful for blue-collar hiring, where candidates may not always rely on formal job portals or written applications.
The platform currently works in English and Hindi, with Vahan.ai planning to add eight more Indian languages, including Tamil, Telugu and Marathi. This could make it easier for workers to discuss jobs, pay, location and shifts in the language they are most comfortable using.
According to the company, its platform has enabled more than 1.5 million job placements across over 920 cities and connected nearly 50 million Indians with employment opportunities.
Why Vahan.ai chose a smaller AI model
Vahan.ai has moved from a 120-billion-parameter model to Nvidia’s Nemotron 3 Nano, which has around 30 billion parameters, making it roughly one-fourth the size. Parameters are the internal settings that help an AI model understand and generate responses.
While larger models can handle complex tasks, a recruitment assistant has a narrower role. It needs to understand candidates, ask relevant questions, answer basic queries and help move the hiring process forward.
That makes a smaller, specialised model a practical fit. It can be faster to run and less expensive while still performing well when trained for a specific task. Vahan.ai fine-tuned Nemotron 3 Nano using its own production data, including real recruitment conversations corrected by humans.
The company tested the model on seven recruitment-specific benchmarks covering response correctness, function calling, human-like interaction, language matching and tool-argument accuracy.
Vahan.ai said the fine-tuned model performed better than both the base Nemotron model and its earlier 120-billion-parameter cloud-hosted model.
Festive season hiring expected to generate up to 2.7 lakh gig jobs: Report
Faster responses, lower costs
Vahan.ai reported nearly 6.7x faster time-to-first response and more than 3x lower average end-to-end response latency. Madhav Krishna, founder and CEO of Vahan.ai, said costs could come down by 60% to 70% or more. The fine-tuning work took around two to three weeks.
Speed matters in blue-collar hiring, where candidates may apply for several opportunities at once. A delayed response can mean losing a candidate to another employer. A faster AI recruiter could help companies engage workers sooner while keeping recruitment costs under control.
A shift towards specialised AI
Vahan.ai’s work with Nvidia shows that, instead of using very large models for every task, companies are exploring smaller models trained for specific business needs.
This could be particularly relevant for India. Smaller models that run efficiently and understand local languages and contexts could help businesses deploy AI at scale without excessive computing costs.