Sigma AI Labs: A recap of our summer 2026 intern program

Sigma AI Labs’ summer 2026 intern program explored four approaches to making enterprise AI systems more capable, efficient, and reliable. One project developed task-oriented agent memory that stores reusable SQL, business definitions, and corrections, improving execution accuracy by 5–10 points while reducing latency and enabling smaller models to compete at lower cost. Another introduced TabularML, which translates natural-language business questions into calibrated classification or forecasting predictions, reaching about 93% of state-of-the-art RelBench performance while operating roughly 100 times faster by avoiding per-task fine-tuning. A third project enabled autonomous planners to derive reusable skills from prior execution traces, improving performance on underspecified text-to-SQL tasks while lowering latency, particularly through skills for revising rejected answers. The final project used a dynamically built concept graph to guide multi-agent deep research across structured and unstructured data, improving grounded insight recall by 11% over the next-best closed-source system at comparable latency and about one-tenth the cost. Together, the projects support Sigma’s goal of delivering assistants that learn from experience, handle ambiguous analytical questions, generate inspectable predictions, and produce evidence-backed insights for business users.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.