Zetagrade

ZetaGrade is an applied AI lab and product studio focused on hard algorithm and data problems, not just wiring together existing frameworks or calling generic LLM APIs. We design and build the core ML/AI systems that sit underneath serious products- ranking and recommendation- retrieval and search- document understanding and knowledge graphs- probabilistic and decision models- domain-specific LLM and agent systems. Our work often starts where “just use an off-the-shelf model” stops making sense. Deep-tech focus- Ranking & retrieval: • Multi-stage retrieval and learning-to-rank pipelines over 100M+ entities (users, posts, candidate profiles, documents) • Embedding-based architectures and sequence models for feeds, search and recommendations - Document understanding & knowledge graphs: • Document-to-knowledge-graph conversion using beam search, constraint satisfaction and entity-linking • Normalising heterogeneous documents (clinical, operational, business) into structured schemas for downstream analytics and automation - Probabilistic & decision models: • Probabilistic graphical models for behaviour and risk (e.g. drug adherence, credit, operational risk) • Throughput and scheduling optimisation using queueing models, simulation and constrained optimisation - LLMs, agents & automation: • Domain-specific LLMs and representation models for search, Q&A and decision support • Retrieval-augmented generation with custom retrieval/ranking layers and evaluation loops • Code and ETL generation models (SAS/Python/R) for data pipelines and validation • Agent platforms with secure data access, classification and policy-aware tool orchestration We usually work with teams who- Have complex, high-stakes data (clinical trials, large talent graphs, financial/operational systems)- Need custom algorithms and models, not just a thin layer on top of someone else’s API - Care about correctness, robustness and long-term maintainability of their AI