Research Intelligence

Emerging Tech Signals

Technologies tracked from arXiv, GitHub trending, and funding announcements by our Research Intelligence department — before they appear in job descriptions.

Frontier Models

↑ accelerating

Mixture-of-Experts architectures gaining traction in production deployments

LLM

3–6 months to job markethigh impact

Test-time compute scaling emerging as training-time compute alternative

Inference

6–12 months to adoptionhigh impact

Multimodal models (vision + audio + text) becoming baseline expectation

Multimodal

Now — already in job specsmedium impact

ML Engineering

↑ rising

vLLM and PagedAttention driving ML infrastructure engineer demand

Inference

Now — active job requirementhigh impact

Model quantisation (GPTQ, AWQ, GGUF) becoming core MLOps skill

Quantisation

3–6 monthsmedium impact

Speculative decoding and KV cache optimisation entering job specs

Optimisation

6 monthsmedium impact

AI Safety & Alignment

↑ fast-growing

Interpretability and mechanistic analysis skills in high demand (UK AI Safety Institute)

Safety

Now — active hiringhigh impact

Constitutional AI and RLHF fine-tuning skills spreading beyond frontier labs

Alignment

6–12 monthshigh impact

Red-teaming and adversarial evaluation becoming standalone job category

Red-teaming

12 monthsmedium impact

Data & Infrastructure

→ stable

Feature stores (Feast, Hopsworks) consolidating as MLOps standard

Data

Established — growing adoptionmedium impact

dbt + Spark + Iceberg replacing legacy data warehouse patterns in AI shops

Data Eng

Nowmedium impact

Streaming ML with Flink/Kafka gaining traction in real-time recommendation

Streaming

12–18 monthslow impact

High-Impact Papers → Job Market Relevance

Theory15k+ citations

Scaling Laws for Neural Language Models

Foundation for all LLM hiring

Systems2.1k citations

FlashAttention-3: Fast and Accurate Attention with Asynchrony

Core ML infra skill

Safety3.8k citations

Constitutional AI: Harmlessness from AI Feedback

RLHF/alignment jobs

Fine-tuning12k+ citations

LoRA: Low-Rank Adaptation of Large Language Models

Standard fine-tuning approach

Architecture4.2k citations

Mixtral of Experts

MoE architecture jobs rising

Alignment3.1k citations

Direct Preference Optimisation (DPO)

Replaces RLHF in many pipelines

Papers surfaced by arXiv Monitor Agent. Relevance scored by semantic similarity to current UK job descriptions. Updated weekly.