AI Builder · Product Engineer · Safety Researcher
Scaling IdeasInto AI Systems.
I design, build, and scale AI infrastructure end-to-end — from first prototype to platforms running at enterprise scale. My focus is on systems that are fast, safe, and genuinely useful.
My craft spans RAG architectures, Constitutional AI, multi-model safety evaluation, and LLM tooling optimized for production — currently pushing the frontier with protocol-level safety via MCP.
Case Studies.
Enterprise AI products, platform orchestration, and scalable systems — built at Microsoft Azure, Wipro, and GovTech scale.
Enterprise SaaS · Microsoft Azure
QaaS 2.5 Orchestration
Consolidated 5 fragmented testing services into a unified orchestration layer serving 500+ engineers at Microsoft Azure — fastest adoption in team history.
Wipro · Embedded at Microsoft Azure
AI Juno & Vega Labs Portal
Unified developer portal merging LLM-powered AI Juno assistant with Vega Labs lab scheduling — serving 12,000+ Azure engineers globally.
DEXter
MCP_SUITE
v1.0 · Protocol Layer
Unified Model Context Protocol interface bridging DEXter Rise, agentic workflows, and system backends — with constitutional safety constraints at protocol level.
// Initialize MCP v1.0 Handshake
const mcp = new DEXterMCP({ version: '1.0' });
mcp.connect({ backend: 'dexter-rise' });
mcp.applySafety({ alignment: 'Constitutional' });
✓ Protocol handshake complete
// Engineering in progress...
▋Research Archive.
Documentation of independent studies on LLM safety, tokenization divergence, and AI alignment.
Tokenization Divergence Research
Identifying vulnerability patterns across Claude, GPT-4, and Gemini; documentation of token-level inconsistency correlations with model safety bypass susceptibility.
Multi-Model Safety Evaluation
Systematic surface analysis of safety metrics and behavioral guardrails across five major model families using API and MCP integrations.
Constitutional AI Assistant Design
Building modular, steerable development assistants grounded in Constitutional AI principles for safe multi-turn reasoning and tool usage.