Applied AI Engineer

Nestium; Union (NJ); On-site; USD 120,000 - 160,000; Full time

Job summary

Nestium is seeking an engineer to design and ship applied LLM systems for extraction, structuring, retrieval, and draft generation with strong grounding and provenance. You will build end-to-end pipelines that output trustable AI results, with emphasis on cited sources and measurable quality.

You will work with data engineers to ensure outputs land in the same provenance and confidence model as other Nestium figures, focusing on production readiness, cost, latency and reliability.

Qualifications

• Shipped LLM or ML features to production (not just notebooks).
• Strong Python skills; comfortable with APIs and data contracts.
• Experience with production debugging and monitoring.

Responsibilities

• Build production LLM pipelines for extraction, classification, summarisation and draft report sections, grounded in cited sources.
• Design retrieval and context assembly over structured market data, registries and geospatial layers (RAG done properly).
• Define evals, golden sets and regression checks to prevent quiet degradation of quality.
• Instrument cost, latency and failure modes; keep human review in the loop for client-facing outputs.
• Collaborate with data engineers and analysts to ensure provenance and confidence in Nestium figures.

Skills

LLM in production

Python

APIs

Data contracts

Grounding

Job description

Nestium uses AI to accelerate real estate intelligence, not to invent it. You'll design and ship applied LLM systems: extraction from messy public sources, structuring into our data model, retrieval over market and geospatial context, and draft generation that a reviewer can trust. Eval, grounding and provenance are first-class — every AI output must be checkable before it reaches a client.

What you'll do

• Build production LLM pipelines for extraction, classification, summarisation and draft report sections — grounded in cited sources.
• Design retrieval and context assembly over structured market data, registries and geospatial layers (RAG done properly, not as a demo).
• Define evals, golden sets and regression checks so model or prompt changes don't silently degrade quality.
• Instrument cost, latency and failure modes; keep human review in the loop wherever the output informs a client decision.
• Work with data engineers and analysts so AI outputs land in the same provenance and confidence model as every other Nestium figure.

What we're looking for

• You've shipped LLM or ML features to production — not just notebooks or prototypes.
• Strong Python; comfortable with APIs, data contracts and production debugging.
• You obsess over grounding, citation and failure cases; you don't trust a fluent answer without evidence.
• You can explain trade-offs (quality vs. cost vs. latency) to engineers and domain experts alike.