Director, AI Engineering & Innovation
Technology executive with a Big 4 consulting background and deep financial-services execution. International career spanning India, UK, Sweden, and 14+ years in Canada — scaling engineering organizations, architecting multi-million-dollar cloud migrations, and pioneering the governed adoption of AI.
By day I lead an AI Engineering & Innovation team at one of Canada's largest banks — agentic systems, LLM platforms, and the governance that lets a regulated business actually use them. Before that I ran the bank's wealth-management data and reporting organization through its move to the cloud, spent seven years in Big 4 consulting leading data and automation practices, and started out building BI and ETL systems for banks and manufacturers across India, the UK, Sweden and Canada.
By night I build and run Lumen, a self-hosted family AI operating system — a Chief of Staff on Telegram, long-term memory, and a bedtime-story studio with its own author, composer and sound designer. Its memory layer became Smara, a product. Everything on this page is something I use.
The intersection of enterprise architecture discipline and hands-on AI product development. I don't just design systems on whiteboards — I build and operate them.
Deep domain expertise across financial services and technology consulting, with product ventures in AI infrastructure, education, and digital media.
Production-hardened systems running on real infrastructure, serving real users. Not demos. Not tutorials. Shipped products.
Family AI operating system with 10 specialized agents, 4-layer memory architecture (Redis, pgvector, Neo4j, Ebbinghaus decay), and multi-channel communication via WhatsApp, Telegram, and web dashboard. Local-first with Ollama, cloud fallback chain.
Memory-as-a-Service API for AI agents. Temporal Memory Scoring with Ebbinghaus decay curves. SDKs for Python, JS/TS, MCP server, LangChain, and LlamaIndex. 96% cheaper than competitors.
58 single-file browser tools — JSON formatter, regex tester, cron builder,
kanban, habits, color palette, markdown editor, encoder, chmod, IP tools, more.
MIT-licensed, no signup, no tracking, your data stays in localStorage.
Built in one night to fill the gaps in a daily workflow.
AI project estimation platform. Upload a BRD/SOW → 9 AI agents run a visual drag-and-drop pipeline → complete project estimate in under 2 minutes.
AI bedtime story production pipeline. Full screenplay format (Pixar/Ghibli quality), 14 TTS voices, 8-dimension quality scoring, visual pipeline IDE.
MCP server compliance testing and monitoring. 50+ automated checks against the official MCP spec. A-F grading system, CI/CD ready.
Covered call ETF investment guide. 5-tab dashboard with strategies, ETF reference, wisdom library, portfolio builder, and DRIP calculator.
Lumen's memory architecture is the core IP. Four layers, each solving a different retrieval problem, unified into a single context pipeline.
The toolchain in active use across a Canadian bank — the platforms, custodians, and legacy systems I work across daily.
Not technologies I've studied. Technologies running in production right now, serving real workloads on parallelromb.dev products.
Architecture is the art of making the right tradeoffs visible. Good systems reveal their intent through structure — the best documentation is the code itself.
The measure of an architect isn't the complexity they can build — it's the complexity they can eliminate while preserving capability.
Building at the intersection of enterprise architecture and AI product development, with a clear trajectory toward executive leadership and entrepreneurship.
Interested in working together? I'm open to architecture & cloud leadership, functional consulting, and AI product partnerships.
Technical deep-dives, product decisions, and lessons from building in public.
An honest comparison of Mem0, Zep, Letta, and Smara with verified pricing — including the tradeoffs each one makes that the marketing pages don't admit.
I shipped a product to the graveyard before it shipped to customers. The story of MCP Doctor and the four-minute validation habit I've made permanent.
The math behind Smara's decay-scored memory — why forgetting is a feature, not a bug.
Community post on building a memory API that works across LangChain, CrewAI, Claude, and any MCP client.