Hrishi Olickel challenges the idea that one universal enterprise ontology can make all information useful. He distinguishes the context of the source, the constraints of storage, the decision being made and the assumptions brought by a model. A customer can mean different things to billing, support and compliance, so he proposes constructing the needed structure when the decision is known rather than forcing every use into one schema.
Hrishi Olickel separates models that design pipelines, inexpensive specialized models that perform repeated work and agents that iteratively optimize the system. Using Southbridge's Ohio campaign-finance entity-resolution project, he describes small development sets, repeated clarification of definitions, second opinions and retained evidence for each answer. The reported cost reductions belong to this particular project, not a guarantee that the same architecture or model will generalize to other datasets.
Hrishi Olickel shows why agreement between models is not sufficient validation: shared assumptions can lose date fields, misclassify contributions or remove meaningful negative signs. He proposes intersecting different model families' judgments while checking persistent shared failure modes. Stable deployment interfaces, bounded cost and access, reusable methodology and explicit failure surfaces support small pilots, versioned checkpoints and gradual optimization of verified production pipelines.
Watch on YouTube
