I bridge business imperatives with modern AI engineering, Model Context Protocol (MCP) gateways, and hybrid retrieval platforms. My focus is taking Generative AI and autonomous agents from fragile prototypes to secure, governed production at Tier-1 banking scale.
Recent articles on enterprise AI, engineering leadership, and technology economics, published on Medium and LinkedIn.
Your AI Agent Has a Budget. But Who Is Accountable When It Overspends? Imagine giving an AI agent a task. It calls a model. Then a tool. The result isn’t right, so it retries. It retrieves more dat…
AI Is Moving From Chatbots to Everyday Work Artificial intelligence is no longer limited to answering questions or generating images. The latest developments show a clear shift towards AI that can …
The AI Unit Economics Problem: Why Cost per Token Is the Wrong Executive Metric Enterprise AI has a measurement problem. We are becoming increasingly sophisticated at measuring tokens, API calls, m…
For a while, most conversations about AI cost started with tokens. How many did we use? Which model was expensive? Why did the bill move this month? Those questions still matter. But they are quick…
The Enterprise Rollout Playbook A practical, verified guide for engineering teams shipping AI coding agents at scale — including routing through Amazon Bedrock, Portkey, or your own in-house LLM. W…
The Enterprise Generative AI Paradox: Bridging the Divide Between Adoption Saturation and P&L Value Realisation Macro-Economic Context and the Enterprise Productivity Disconnect The global enterpri…
Five working AI agent skills connecting source-code graphs, architectural decisions and durable Obsidian memory across Codex, Claude and MCP.
Long-form perspectives on enterprise AI engineering, hybrid RAG, agentic reliability, and technology economics.
Why routing every query to frontier models is architectural negligence: implementing intent-aware routing between on-prem small models and frontier LLMs with automatic circuit-breakers.
Deconstructing the 5 perimeter failure modes—PII leakage, semantic cache drift, unbudgeted token egress, and permission desync—that kill enterprise GenAI pilots before production.
Deconstructing prompt token bloat, cache miss penalties, reranker latency, and egress costs: an engineering guide to modeling enterprise LLM unit economics.
Reference architectures and security topologies developed to bridge the gap between AI proofs-of-concept and scalable enterprise production.
A battle-tested production blueprint for enterprise search and knowledge retrieval combining dense semantic embeddings, sparse BM25 indexing, and cross-encoder reranking.
Multi-hop query decomposition, dynamic query reformulation, and autonomous citation critique for high-complexity enterprise research.
A unified reverse-proxy platform layer enforcing security, multi-provider model routing, semantic caching, token quotas, and audit logging across all enterprise apps.
Defense-in-depth security framework protecting production AI systems from indirect prompt injection, data exfiltration, jailbreaks, and adversarial poisoning.
Bhavin's curated technology tracking across Adopt, Trial, Assess, and Watch categories based on real-world enterprise viability.
Automated regression testing of prompt templates, RAG ground truth, and reasoning chains in pull requests using golden SME benchmarks.
Standardized open protocol connecting AI agents securely to enterprise data sources (SQL, GitHub, Jira, Obsidian) with granular RBAC.
Dynamic task complexity classifiers routing 70% of routine intent/extraction to low-cost small models (Llama 3.3 / Mistral) and 30% to frontier LLMs.
Augmenting vector embeddings with knowledge graph entity-relation trees (Neo4j / Graphiti) to resolve multi-document corporate relationships.
Vision-guided autonomous agents interacting directly with legacy desktop and web applications via keyboard/mouse emulation.
Multi-agent orchestration architectures (e.g. LangGraph supervisor-worker networks) coordinating specialised agent roles.
Autonomous software agents authorized to execute programmatic financial transfers, invoices, or settlements.
End-to-end agents writing code, merging PRs, and deploying directly to customer-facing production infrastructure without human sign-off.
Transparent, client-side engineering and financial calculators with zero black-box assumptions.
Model monthly vector storage, embedding tokens, inference volume, and caching return across enterprise document pools.
Compute daily, monthly, and annualized inference budgets based on prompt/completion ratios and prompt cache hits.
Map candidate AI initiatives into Quick Wins, Strategic Bets, Experiments, or Defer based on business value and risk.
A rigorous 50-point engineering, security, and FinOps verification gate before promoting Generative AI and Agentic systems to live enterprise traffic.
Senior Engineering Manager at Commonwealth Bank, with a career spanning AI strategy, platform engineering, architecture, and globally distributed delivery.
I work with senior leaders, product teams, and engineers to turn business priorities into delivery roadmaps and dependable services. My approach brings engineering, architecture, and domain expertise together early, with quality and governance built into the operating model.