Agentic AI
Capital projects generates the same bottleneck — dozens of vendor proposals, hundreds of technical and commercial criteria, and a procurement person manually cross-referencing all of it under deadline. We're building an AI tool that reads bid packages the way an experienced procurement specialist does: comparing proposals against your evaluation criteria, flagging commercial exceptions and technical gaps, and citing exactly where in each proposal it found the answer.
No black box, no invented data — every output traces back to a real page in a real bid.
We're testing it against real packages from our own mining and EPCM project work before it goes anywhere near a client deliverable. Small scope, real accountability, built by people who evaluate bids for a living.
More to come as it takes shape — including how this fits into the broader toolset we're building under Glintz AI.
RAG
Retrieval-Augmented Generation: searching our own procurement data first, then giving those results to the AI to write an accurate, grounded answer, before accessing the LLM model.
One agent, multiple subagents
One agent manages multiples subagents that do their specialized work: one compares the material price, one compares the manpower cost, one evaluates the transports dynamic cost. Teamwork.
Guardrails
We use guardrails: rules and checks that keep AI answers accurate, safe, and on-topic — catching errors before they reach you. Minimize or even eliminate the ''hallucinations''.