About the Company:

NewStreet is an award-winning fintech building AI-powered platforms that transform how financial institutions operate. Established in 2017 and ISO 9001 & ISO 27001certified, we serve 170,000+ customers across lending, trade finance and payments, partnering with leading banks across India, the GCC and Southeast Asia. Financial Times - 7th Fastest-Growing Fintech, Asia Pacific - Deloitte Technology Fast 50 - Aegis, Graham Bell Award - World Fintech Awards 2025 - FICCI IBA Best Emerging Technology Platform

About the Role:

MiFiX.ai is our AI platform for financial institutions - a set of engines and agents that read documents, answer questions over bank data, run rule-driven decisioning and hold a conversation with an end customer. It is in production with real banks.

This role owns the model side of it.

You run the AI / ML function: the models we fine-tune, the retrieval pipelines behind them, the evaluation harness that says whether a change helped, and the cost and latency of everything we serve. You have one AI / ML Engineer and a fellowship bench of ten fellows to bring up - a large part of this job is turning that bench into people who ship.

You report to the Head of Engineering and sit alongside the platform, security and DevOps functions. This is a hands-on leadership role: you are expected to be in the code and in the eval numbers, not only in the review.

Key Responsibilities:

What You Bring

- 5-8 years in machine learning, with at least 2 years on systems that served live

traffic - not research-only.

- Strong Python and PyTorch. You have run a fine-tune end to end and can explain

which metric moved and why.

- Production RAG built by you: chunking strategy, embedding choice, reranking,

and the evaluation that proved it worked.

- An evaluation harness you designed - labelled data, regression gates, a number

you were willing to be held to.

- Demonstrated control of inference cost and latency, with before-and-after

figures.

- Experience mentoring early-career engineers, and the willingness to give direct

feedback.

- BE / BTech / MS in Computer Science, Mathematics, Statistics or a related

quantitative field.

What Would Be Great to Have:

- OCR and document-extraction pipelines on poor-quality scans.

- Speech - ASR, diarisation or TTS in production.

- ISO 42001, NIST AI RMF or equivalent model-governance exposure.

- On-premise or air-gapped model deployment for a regulated client.

- Financial-services domain: lending, collections, trade finance or onboarding.

- Open-source contribution, publication or a public model/dataset release.

Tech Stack

- Models: Open-weight LLMs, PyTorch, Hugging Face, LoRA / QLoRA, quantisation

- Retrieval: Vector search, hybrid retrieval, rerankers, structured grounding

- Serving: Python, FastAPI, containerised inference, GPU and CPU deployment

- Evaluation: Custom harness, labelled regression sets, CI gating

- Governance: Model cards, lineage, ISO 42001 control mapping