Member of Technical Staff - Applied AI Research

Vaya; Delhi; On-site; INR 1,500,000 - 2,500,000; Full time

Job summary

Vaya is seeking an AI researcher to enhance language models for Hindi and other Indic languages. Your role includes fine-tuning models to improve fluency and accuracy, designing retrieval strategies, and managing evaluation processes to measure model performance.

The ideal candidate will possess strong engineering skills and a deep understanding of language processing challenges. Join us in pushing the boundaries of conversational AI at our office in Vasant Kunj, Delhi.

Qualifications

• Experience with LLM fine-tuning and managing datasets.
• Understanding of tokenization and code-switching in Indic languages.
• Strong engineering skills in production environments.

Responsibilities

• Fine-tune models for Hindi and other Indic languages.
• Build and manage fine-tuning pipelines and evaluation.
• Collaborate to measure quality effectively.

Skills

LLM fine-tuning

Indic NLP challenges understanding

Engineering skills

Experimental rigor

Tools

Elixir

TypeScript/Bun

PostgreSQL

NATS

Job description

About Anuvaya

We think conversational AI agents will deliver all professional services in India. We started with astrology. We are a small group of engineers, designers, and product folks building at the intersection of conversational AI and domain expertise. Making an AI agent sound human‑like is hard. Making an AI an expert in a domain is also hard. We’re doing both together. We’re backed by Accel, Arkam Ventures, and Weekend Fund.

The Role

Frontier models are incredible at English, but are not as good at Indic languages, and our users speak Hindi, Hinglish, Tamil, Telugu, and many others. Our agent needs to be fluent, accurate, and domain‑expert level across all of them.

The Applied AI Research team owns two problems. First, fine‑tuning models for our use case to improve accuracy, fluency, and domain understanding in Indic languages where foundation models fall short. Second, making those models work reliably in production through prompt engineering, context management, retrieval strategies, and systems that turn a capable model into a domain expert.

You’ll bridge the gap between research and production. Some weeks you’ll run fine‑tuning experiments to improve Hindi response quality. Other weeks you’ll redesign how context flows through a multi‑turn conversation. The through‑line is the same: make the model better at doing what our users need, in the language they think in.

What You’ll Do

• Fine‑tune models for Indic language performance, improving fluency, accuracy, and domain understanding in Hindi, Hinglish, and other Indian languages
• Build and manage fine‑tuning pipelines, data curation, training runs, evaluation, and deployment of fine‑tuned models
• Work with the team on prompt engineering and context management, designing how the model receives and reasons over information across multi‑turn conversations
• Design retrieval strategies that get the right domain data to the model at the right time
• Run experiments on model behavior to see how fine‑tuning, context structures, prompt formulations, and tool designs affect output quality across languages
• Collaborate with Evaluation to measure what actually matters, especially for subjective, language‑dependent quality
• Stay current with frontier model capabilities and figure out how to exploit new features the day they ship

What We’re Looking For

• Experience with LLM fine‑tuning – trained or fine‑tuned models, managed datasets, and evaluated results
• Understanding of Indic NLP challenges: tokenization, code‑switching (Hinglish), script diversity, and where current models fail
• Experience with LLMs in production: prompt engineering, context management, retrieval‑augmented generation
• A clear understanding of the difference between a demo and a production system and having fought with context windows, hallucinations, and inconsistent model behavior
• Strong engineering skills – you ship code, not just papers. Your research runs in production
• Experimental rigor – you design experiments, control variables, and know when results are significant

Bonus

• Experience fine‑tuning on Indic language data or multilingual corpora
• Experience with Claude, Gemini, or other frontier model APIs at depth
• Familiarity with our stack: Elixir, TypeScript/Bun, PostgreSQL, NATS
• Published work or substantial projects in applied NLP, multilingual models, or knowledge‑grounded generation
• You’ve read our whitepapers (Realtime Context Engine, Context Splitting) and have thoughts on what we got wrong

How We Work

We care about craft obsessively. Your work gets questioned, pulled apart, and rebuilt not because we’re harsh, but because everyone here holds each other to a standard most places don’t bother with. We work out of a hacker house in Vasant Kunj and strongly encourage everyone to be in office.

If that sounds like the only way you want to work, let’s talk.