Please be aware. We don't hire without loads of relevant experience. And secondly, we don't pay European salaries for any offshore job. We can however give you a promissing career with nice bonusses.
Senior AI / LLM Engineer – Legal‑AI Portal (Remote
- India‑based talent welcome)
Please do not respond if you dont have extensive experience. You will waste your own and also our time.
We don't pay european sallaries to everyone working remote!!
About Cognita Innovative Solutions
Cognita is a Dutch‑led, fully remote scale‑up building the Juridical AI Portal—a cloud‑native platform that digests entire lawsuits, runs two specialised LLM “advocates” against each other, and produces judge‑style syntheses, draft pleadings and success‑probability dashboards. Our stack already combines DeepSeek‑VL 1.5+, LLaMA 3 (13 B / 70 B) and Qwen 3.3—each fine‑tuned on millions of Dutch & EU rulings, statutes and law‑review articles—We go live with beta partners in September 2025. Your mission is to push our models, prompts and infra to production‑grade brilliance.
What you’ll design, build & own
Fine‑tune and distil foundation models (DeepSeek, LLaMA 3, Qwen 3.3) with LoRA, QLoRA and continual‑pre‑training on Dutch legal corpora (ECLI, Kamerstukken, journals).Engineer multi‑agent reasoning loops (“Attorney 1 → Attorney 2 → Judge”) in Python / LangChain, optimising for factual consistency, latency (Craft high‑impact prompts & system instructions that balance persuasiveness, rebuttal logic and judicial neutrality—then A / B test with human reviewers and RLHF pipelines.Build retrieval‑augmented generation (RAG) pipelines over OpenSearch / FAISS with Sentence‑Transformers or Legal‑BERT embeddings for 3 M+ rulings and 2.5 M+ articles.Automate eval & monitoring : BLEU / ROUGE / BERTScore, hallucination detection, consistency‑under‑retries, judge‑panel scoring dashboards in Prometheus + Grafana.Deploy production endpoints via SageMaker TGI or LMI, using Docker, GitHub Actions, quantisation (4‑bit AWQ / GPTQ) and tensor parallelism on A100 GPUs.Champion security & compliance : red‑team prompts, PII masking, adversarial testing, and model‑card documentation per ISO / GDPR.Mentor mid‑level ML engineers, codify best practices in Confluence, and keep tech debt near zero through rigorous code reviews.You’ll thrive here if you have
Must‑haves :
6+ years in applied ML / NLP with at least 2 years on LLMs.Deep Python & PyTorch; hands‑on with HuggingFace Transformers / PEFT.Proven track record of fine‑tuning and evaluating models ≥13 B parameters.Expertise in prompt engineering, chain‑of‑thought, RLHF or DPO.Strong grasp of RAG, vector‑stores, embeddings & retrieval latency tuning.Solid MLOps skills : Docker, GitHub Actions, AWS SageMaker / Lambda.Security mindset : guard‑rails, content‑filtering, jail‑break resistance.Two verifiable professional references.Bonus points :
Experience shipping multi‑agent LLM systems in production.Knowledge of Rust or Go for high‑performance inference.Worked with 128 K‑context models (e.g., Qwen, Claude).Research publications in legal‑tech or language modelling.Familiarity with Dutch or EU legal systems.GPU cluster optimisation with Triton / FasterTransformer.What we offer
Competitive, market but local sallary. 1500-3000 (very, very max) USD / MonthPermanent, fully remote position with at least four hours of overlap with CET.A culture built on loyalty, intellectual rigour and purposeful hard work—when we win, we win together.Please save yourself and us the trouble. If you dont have a lot of ML / AI engineering experience, than don't apply. Also don't apply if you think we pay you a western salary for working remote!!!
