Jirav — Rohin Sood

A financial answer needs more than convincing prose. It needs the right records, the right dimensions, and a calculation that can be repeated. At Jirav, I worked on the backend connecting natural-language analysis to those foundations.

Give the agent a useful interface

I architected an MCP layer over more than 15 REST APIs and 50 financial datasets. It translated questions into structured revenue, cost, and departmental retrieval for Jirav’s financial planning and analysis platform.

Make the numbers repeatable

I implemented deterministic retrieval and time-series analysis for revenue and cost-of-goods-sold queries, margin calculations, variance and scenario comparisons, and dimensional slicing. Keeping these operations explicit makes the calculation easier to inspect and reproduce.

Carry the context through

I developed a natural-language engine for SaaS artifacts, connecting business records with predictive insights and contextual analysis across customer, product, and department dimensions. The same number can mean different things depending on which period and slice of the business it represents.

What this work sharpened

This experience deepened my interest in tool orchestration and backend infrastructure. The interface between a model and its data is where a broad question becomes a precise operation—and where a useful AI application earns its credibility.