ThirdLaw Molecular

• Build the evaluation layer in the ThirdLaw platform for LLM prompts and responses • Design and tune guardrails, classifiers, and semantic judgment systems in real-time • Implement evaluation strategies with semantic similarity, foundation model scoring, and rule-based systems • Integrate model outputs with downstream enforcement actions (e.g. redaction, escalation, blocking) • Prototype, tune, and productize small language models for classification, labeling, or scoring • Collaborate with data infrastructure engineers to connect evaluation logic with ingestion and storage • Build tools to observe, debug, and improve evaluator performance across data distributions • Define abstractions for reusable evaluation components that can scale across use cases

• Architect and evolve core backend systems—event ingestion, real-time risk evaluation, semantic search, and investigation-grade querying. • Design and optimize streaming data pipelines using Kafka, Redpanda, or similar technologies to support high-throughput, low-latency processing of LLM interactions. • Define and refine data models and schemas across structured (Postgres), analytical (ClickHouse), unstructured (S3), and vector (Qdrant) storage layers. • Identify architectural weaknesses and performance bottlenecks across ingest, indexing, and query layers; lead improvements to support scalability, correctness, and observability. • Implement data ingestion and transformation frameworks that support policy evaluation, semantic analysis, and dynamic response. • Optimize query performance, storage efficiency, and retrieval accuracy to support deep investigation and real-time enforcement. • Build and maintain backend APIs, monitoring infrastructure, and policy enforcement workflows that underpin the ThirdLaw platform.

• You’ll work within a small but mighty team of AI engineers and backend engineers to build and maintain application features for AI-native platform. • Every day, you will build new and innovative product features. • Start-up responsibility; you are the first and often last stop on whether a feature is good or great.