Staff AI Engineer (New York)

Xenoss is an AI engineering and integration services company, helping medium to large enterprises run AI transformation end-to-end, from situation analysis and goals framing to data discovery and preparation, pipeline building, model development, retraining pipeline design, solution deployment, and support.

We build a broad spectrum of AI solutions such as user behaviour prediction, content generation, NLP, audience segmentation, pathfinding solutions, AI assistants, edge computer vision, fraud detection, and others.

We work with prominent companies such as Microsoft, Toshiba, AstraZeneca, Activision Blizzard, Verve Group, Voodoo Games, and Telefonica, among others.

We’re included in the top 100 software companies on the Inc. 5000 list.

About the Role

We’re hiring a Staff AI Engineer to lead fine-tuning and domain adaptation of large language models on top of one of the most complex enterprise datasets you can work with: a multi-year archive of real customer conversations from a world-leading banking holding.

This role sits at the intersection of speech, language, and operational decision systems. The objective is not generic chatbot improvement. The objective is to turn raw audio and transcripts into production-grade intelligence: classification, intent detection, risk signals, quality insights, and agent copilots.

You will own how these models are trained, evaluated, and operationalised.

What You Will Do

You’ll operate across the full fine-tuning lifecycle from dataset engineering to model deployment.

You’ll work with large volumes of audio and transcript data, transforming unstructured conversational artefacts into structured instruction datasets suitable for supervised and alignment training.

Core work includes:

• Designing fine-tuning strategies for conversational financial data
• Structuring transcript corpora into task-ready training formats
• Running LoRA / QLoRA training pipelines on open-weight LLMs
• Defining evaluation frameworks and quality benchmarks
• Leading structured error analysis and iteration cycles
• Optimizing models for latency, cost, and deployment constraints
• Partnering with MLOps on serving and monitoring

You’re expected to be deeply hands-on in training infrastructure and experimentation.

This is not an oversight-only Staff role.

Technology Landscape

You’ll operate within the modern open-model fine-tuning ecosystem, including, but not limited to:

• Open-weight LLMs (LLaMA, Mistral class)
• Parameter-efficient training (LoRA / QLoRA)
• Alignment optimization where relevant
• PyTorch training pipelines
• HF ecosystem (Transformers, TRL, PEFT)
• Quantization and optimized inference runtimes

We optimize for production viability, not academic benchmarks.

Scope of Ownership and Delivery Context

At Staff level, you’ll own both the fine-tuning architecture and its execution across the full lifecycle, from dataset engineering through production deployment.

Core Ownership

• Define fine-tuning and iteration strategies
• Establish evaluation frameworks and acceptance criteria
• Drive trade-offs between model quality, cost, and latency
• Act as an escalation point for performance and architecture decisions

Team and Delivery Context

• Work within a cross-functional team spanning AI engineering, MLOps, data engineering, and client stakeholders
• Mentor engineers running training pipelines
• Partner with domain SMEs on labelling frameworks

Requirements

• Hands-on LLM fine-tuning
• Experience with conversational or speech-derived corpora
• Deep familiarity with LoRA / QLoRA and PEFT methods
• Ability to design evaluation frameworks, not just run them
• Comfort working with messy enterprise data
• Experience deploying models into production stacks

Nice to Have

• Financial services domain exposure
• Speech / ASR pipeline familiarity
• Model governance and auditability experience

Looking for Another Position?

See all our open positions and learn why you should consider joining the Xenoss team.

Careers at Xenoss