About the role

Hunar.ai is hiring a Research Engineer to turn the latest AI advances, especially in voice AI, into faster and cost-effective systems for our product. You will track what is new, decide what matters for Hunar.ai, and prove or disprove it with quick proofs of concept.

A major focus is optimising the performance and cost of our voice AI pipeline using open-source models. Typical questions you might answer: which caching techniques cut latency and spend the most, and how our decision models should work. Speed matters here: we want ideas validated or discarded in days, not months.

What you'll do

- Stay on top of advances in AI relevant to Hunar.ai, especially voice AI: new ASR, LLM, TTS and speech-to-speech models, papers, inference techniques and tooling.
- Filter the noise: assess which advances could improve our product, and share concise write-ups and recommendations with the team.
- Optimise the performance and cost of our voice AI pipeline using open-source models, targeting latency, quality and cost per conversation.
- Research and design how our decision models should work, for example routing between models, turn-taking and intent decisions, and when to use a small versus a large model.
- Benchmark open-source models against commercial APIs on accuracy, latency and cost, using realistic test sets from our domain.
- Run fast proofs of concept to validate or discard ideas, with clear success criteria and measured results.
- Hand off successful POCs to the engineering team with enough documentation and code to take them to production.

What we're looking for
- 2+ years of experience in ML engineering, applied research or a closely related role.




- A track record of reading research papers and turning them into working code quickly.
- Hands-on experience running and evaluating open-source models (for example Llama, Qwen, Mistral, Whisper or open TTS models) on your own GPU setup.
- Experience with model serving frameworks such as vLLM, SGLang, TensorRT-LLM or Triton.
- Rigorous experimental habits: clear hypotheses, sound benchmarks, honest reporting of negative results.
- A bias for speed: you can scope a POC, build it in days and decide whether to continue or stop.
- Clear communication, so non-researchers can act on your findings.

Nice to have

- Experience with voice AI pipelines in production: ASR, TTS, voice activity detection, turn-taking or speech-to-speech models.
- Solid understanding of inference optimisation: quantisation, batching, KV caching, speculative decoding and streaming inference.
- Experience fine-tuning or distilling models, including LoRA and other parameter-efficient methods.
- Experience designing routing or cascade systems that pick the right model per request.
- Familiarity with evaluation of speech systems, such as word error rate, MOS and latency metrics like time to first audio.
- Experience with multilingual or Indic-language speech and text models.
- Publications, open-source contributions or public write-ups on ML topics.

Role details Location - Bangalore

Employment type - Full-time

Experience - 2+ years

Apply here: https://apply.hunar.ai/job/35jkpftr?utm=jq_af

You can aslo send your CV and a link to a POC, write-up or repo that shows how you test ideas quickly to [email protected]

#OpenSourceAI #LLMInference #InferenceOptimization #AppliedAI #BangaloreJobs

📌 Research Engineer (Bengaluru)
🏢 Hunar.AI
📍 Bengaluru