Senior AI/ML Engineer (LLM)

Job Description:

Don’t Apply if you are a Fresher

Senior AI/ML Engineer (LLM) — Nabo.E

About Nabo.E

Nabo.E builds conversational AI agents and customer intelligence for businesses that serve customers at scale. Our agents handle voice, WhatsApp, Instagram, SMS, email and web chat end to end, from first enquiry to confirmed booking, and turn every conversation into intelligence that grows revenue.

We are in production with government bodies and fast-growing businesses in India. . The models behind this are the product, and we are hiring the engineers who will build the next generation of them.

The role

Title: Senior AI/ML Engineer (LLM) Experience: 5+ years

Location: Hyderabad Type: Full-time

You will own the training, fine-tuning and serving of the large language models that power Nabo.E’s agents. This is a hands-on engineering role: you will write the training code, run it on multi-GPU clusters, evaluate the results on real customer conversations and ship models to production across six channels in multiple Indian languages.

You report to [Head of AI / CTO] and work directly with the founders, the platform team and client-facing engineers.

What you’ll do

* Design, pretrain and fine-tune transformer language models at 30B+ parameter scale, from data pipeline to checkpoint.

* Run distributed training across multi-node GPU clusters using FSDP, DeepSpeed or Megatron-LM; profile and fix throughput, memory and stability issues.

* Build and maintain data pipelines for large-scale text and conversation corpora: deduplication, filtering, tokenization, quality scoring, multilingual (Indian language) coverage.

* Post-train models for Nabo.E’s agents: SFT, DPO/RLHF, tool use, function calling and structured output on real customer conversations.

* Build evaluation suites that measure what matters to us: intent classification accuracy, booking conversion, hallucination rate, latency, multilingual quality.

* Optimise models for production serving: quantization, distillation, speculative decoding, vLLM/TensorRT-LLM deployment, cost per conversation.

* Train the predictive models behind our playbooks: booking likelihood, drop-off risk, best time and channel, add-on fit.

* Write clean, tested, reviewable Python; own your code from experiment to production.

* Mentor other engineers and set the technical bar for ML work at Nabo.E.

Must have

* 5+ years of hands-on ML/AI engineering, with at least 2 years on large language models.

* Conversational intelligence experience. You have built or worked on a conversational AI or customer-intelligence platform in production: chatbots, voice agents, contact-centre AI, intent classification, dialogue management, conversation analytics. You understand what breaks when real customers talk to a model at scale.

* Python expert. Deep understanding of the language itself: the data model, memory and the GIL, async, generators, typing, metaprogramming, packaging and profiling. You write idiomatic, performant code and libraries others depend on. We will test this.

* Strong algorithms and data structures. You reason about complexity and choose the right structure under real constraints, and you can solve hard problems on a whiteboard and in code.

* Architectural ability. You have designed systems end to end: data pipelines, training infrastructure, model serving, APIs. You make trade-offs deliberately and can defend them.

* Logical, structured thinking. You break ambiguous problems into clear parts, form hypotheses, and let evidence decide.

* Data science fundamentals. Statistics, experiment design, evaluation metrics, exploratory analysis with pandas/NumPy/SQL; you can pull insight out of millions of conversations.

* Deep PyTorch expertise: custom modules, autograd, mixed precision, torch.compile, CUDA-aware debugging.

* Proven experience training or fine-tuning models at 30B+ parameters, including distributed training across multiple nodes (FSDP, DeepSpeed ZeRO, Megatron-LM, or equivalent). You can explain the parallelism choices you made and why.

* Strong understanding of transformer internals: attention variants, positional encodings, MoE, KV-cache, tokenizers, scaling laws.

* Experience building large-scale data pipelines for pretraining or fine-tuning corpora.

* Experience with post-training methods: SFT, DPO, RLHF, or similar.

* Experience taking models to production: serving, monitoring, cost and latency optimisation.

...