Company description

Finteli is an early-stage, pre-seed startup building B2B licensed software for financial practices. We help firms run structured, AI-assisted work on real client files with qualified accountants staying in control of outputs before anything is finalised.

Today, product features are powered by hosted large language models and structured application logic. Our year-one ML roadmap is to build Finteli-owned, domain-tuned models that reduce reliance on third-party APIs, improve cost predictability, and strengthen data control for professional services customers.

Role description

This is a high-level, applied ML engineering role. You will work with broad autonomy: independently planning how to achieve the goals and projects set by the CEO, then executing against year-one milestones toward Finteli-owned specialist models for core agent and job tasks.

You will ship models and evaluation pipelines that improve real product workflows (bookkeeping review, VAT prep, reporting support), working closely with product engineers - not in a pure research lab. Expect to define training and evaluation datasets, run fine-tuning and alignment experiments, integrate models behind existing job and agent flows, and establish quality bars aligned with accountant correction rates and citation usefulness.

The product today is TypeScript-first (web client, Node API, Postgres). You will deliver models and APIs or containers that integrate cleanly with application-owned orchestration, respect multi-tenant B2B licensed software constraints from day one, and participate in human-in-the-loop UX where ML outputs are drafts until an accountant approves.

Required skills & experience

Strong Python for production ML code (3+ years), not notebook-only workflows
Hands-on experience fine-tuning or adapting LLMs for a specific domain or task family
Solid understanding of transformer-based LLMs - prompting limits, context windows, structured outputs, and failure modes
Experience building evaluation pipelines - held-out sets, rubrics, error analysis, and iteration from user feedback
Experiment discipline - configs, seeds, logging, reproducibility; Git and code review
Familiarity with document-heavy workflows (PDF, Excel, long-form reports) in NLP or information extraction
Ability to collaborate with software engineers shipping TypeScript/Node product code - you communicate clearly across ML and application boundaries
Awareness of data privacy in B2B or regulated contexts (customer data must not leak across tenants or into public training without policy)
Right to work in the UK

Nice to haves

Hugging Face ecosystem - Transformers, PEFT/LoRA, TRL, datasets, tokenisers
PyTorch (primary) or JAX experience at scale
RAG and grounding - chunking, embeddings, citation-style retrieval
Financial or accountancy domain - VAT, bookkeeping, management accounts (UK preferred)
OCR / layout models for scanned PDFs and tables
Quantisation and efficient inference - vLLM, TensorRT-LLM, ONNX, or similar
MLOps - Weights & Biases, MLflow, DVC; CI for model smoke tests
Cloud GPU training and cost budgeting
On-prem / private deployment experience for enterprise customers
Synthetic data generation with human validation loops
Safety / red-teaming for professional advice boundaries

What we offer

£75,000 – £90,000+ salary
25 days holiday allowance + 8 days for bank holidays
Favourable sick pay
Fully remote - we are a fully remote company
Flexible working
Enhanced pension contributions
Regular salary reviews
Performance bonuses
Training allowance
Potential equity (depending on growth and performance)

Apply by email using the Apply button above.