AI systems · Principal’s professional experience
Putting enterprise controls around AI assistance.
An LLM and retrieval-augmented architecture assistant exploring how natural-language requirements can become grounded recommendations and diagrams.
This page describes Dhruv Doshi’s professional experience in prior roles; it is not a client engagement delivered by Arihant Global Ventures Inc.
The context
An architecture assistant must do more than generate a plausible response. Its recommendations need traceable sources, appropriate permissions, evaluation, and human review.
Dhruv’s contribution
- Designing permission-aware retrieval and citation validation for architecture knowledge.
- Building offline evaluations for recommendation quality, grounding, and failure behavior.
- Applying guardrails and approval boundaries before generated output can influence decisions.
- Exploring MCP and agent-gateway patterns for controlled tool access.
What the work enabled
Development work explores how probabilistic model behavior can operate behind deterministic controls. This is not presented as a completed production deployment.
Relevance to your organization
For a prospective AI engagement, the first deliverable should define the system boundary and evaluation criteria—not only a model demonstration.
Explore the related capabilityStart with a conversation.
A free 30-minute introduction to discuss your AI governance challenge, constraints, and whether we can help. No project brief required.
Contact us for current availability. Scope, timing, and fees are agreed before paid work begins.