Technical Solutions Architect, Evals & Fine-Tuning at Innodata Inc.
$140,000–$160,000from the description
Jul 20, 2026
—
Jul 21, 2026
What this job asks for AI summary
A senior individual-contributor role bridging customer-facing technical consulting and AI/ML delivery. The position involves leading discovery with foundation model labs and enterprise AI teams, then designing end-to-end post-training and evaluation solutions — covering fine-tuning pipelines, preference data collection, custom benchmarks, and human-in-the-loop eval workflows. It suits an experienced ML practitioner who has built these systems hands-on and can translate ambiguous client goals into scoped, deliverable engagements.
Senior level · 7+ years · Remote · Full-time
“or” means any one of them counts — you don't need all of them.
We read this from the posting text with AI. Skim the description below before ruling yourself out.
How this req sits in the market our data
Roughly 1,050 people nationally plausibly meet what this posting asks for (computer occupations, all other). range 220–1,550
Applicant volume Moderate — A normal amount of company. The rare requirements below are what will separate a shortlisted application from the rest.
Rare in this occupation — lead with these, and say what you built with them.
Most people in this occupation already list these. Still required — just not what gets you shortlisted.
What the occupation pays Median $119,144 (middle half $81,115–$160,963). This posting is about at that midpoint.
Estimated from BLS employment for this occupation and area, per-skill prevalence across our listing corpus, and published wage benchmarks — as of Jul 28, 2026. It is a model, not a headcount.
Why we read it this way (9)
The role title 'Technical Solutions Architect for Evals & Fine-Tuning' carries no seniority level word, so the title states no level; however, the 7+ years requirement and senior IC framing support a Senior classification.
SOC classification is genuinely ambiguous: the role is a pre-sales/solutions architect for AI/ML products, sitting between 15-1299 (Computer Occupations, All Other — the closest match for ML solutions architects) and 15-2051 (Data Scientists, given the deep LLM eval and fine-tuning hands-on work). 15-1299 was chosen as primary because the dominant day-to-day work is client-facing technical solutioning, proposal authoring, and engagement design rather than original modeling or research.
RLHF, DPO, and KTO are listed together as preference optimization methods ('RLHF, DPO, or KTO strongly preferred') under the requirements section but qualified as 'strongly preferred' rather than hard-gated — treated as preferred with alternatives reflecting the interchangeable options.
lm-evaluation-harness and lighteval are listed as examples of evaluation frameworks ('such as lm-evaluation-harness, lighteval, or equivalents'); they appear in the requirements block and are treated as a hard gate on the capability, with lighteval as an alternative.
vLLM and Hugging Face appear in the requirements block as part of 'the modern LLM toolchain' and are treated as hard gates.
The degree requirement states 'Bachelor's or advanced degree … or equivalent demonstrated experience' — the explicit equivalent-experience escape means no formal degree is hard-required.
The 2+ years specifically on LLM evaluation and/or post-training is a sub-requirement within the 7+ year overall gate; the overall years minimum reflects the top-level 7+ year requirement.
RLAIF, rejection sampling, and distillation are mentioned in the 'stay current' context rather than as a hard gate or preferred qualification — RLAIF is included as preferred given its explicit mention alongside DPO/RLHF in the post-training list; rejection sampling and distillation are generic methodology terms and are omitted per the no-generic-concepts rule.
Caller marked this a fully-remote role — scored against the national candidate pool.
Read the full posting
The employer publishes the full description on their own site — read it there ↗. Or sign in to read it here — it's free, and it also lets you track this application.