I build AI systems.
I've been at this long enough to know where most of them fail.
I'm Rupreet Gujral - an AI and Systems Architect with 25 years spanning enterprise tech, global consulting, and the builder trenches. I've founded startups, shipped patents, and worked inside organisations large enough to know why most AI projects stall. Today I architect agentic systems and LLM infrastructure - and I write about what I learn, without the hype.
Not a ladder. A loop.
"Most careers in tech are a ladder. Mine has been a loop - and it's given me something a straight line never could."
Corporate to startup, practitioner to strategist, founder to architect - and back again. 25 years across enterprise tech, consulting, and building my own ventures has given me a vantage point that's hard to get from one track alone.
I've sat in rooms where AI projects die slow deaths - not because the technology failed, but because the architecture was wrong from day one. I've also been on the other side - founding startups, shipping under pressure, learning what actually works when the demo is over and the budget is real.
Today I architect agentic systems: multi-agent pipelines, LLM infrastructure, cost governance, observability. I write about what I'm building and learning - in real time, from first principles, not from a vendor whitepaper.
From the blog
Practitioner notes on agentic systems, LLM infrastructure, and what I learn building AI in the real world - not the demo version.
Your agent works. That's not the same as production-ready.
I've been in enough post-mortems to spot the pattern before it plays out. A team builds an agent. It handles the happy path well. The demo goes smoothly, the stakeholders nod, someone asks "when can we ship?" And that's where it gets quiet. Because nobody has actually thought through what "ship" means for an agent. It's not like shipping a web app, but the mistake most teams make is assuming it is. With a web app, the path to production is well-trodden: CI/CD, test coverage, error handling, ro
Prompt Caching Isn't a Setting. It's a Prompt Architecture Decision.
How prefix caching actually works, why most teams get the ordering wrong, and the exact restructuring that turns a cache-miss prompt into a cache-hit one. I once reviewed a system where every single call to the model was a cache miss. Not most calls. Every call. The team had read the docs, added the caching parameter where the provider asked for it, and moved on assuming it was handled. It wasn't. Somewhere in their prompt construction, a request ID had been added to the top of the payload "for
RAG Isn't a Vector Database. It's a Retrieval Problem.
Every team building with RAG right now has roughly the same setup. Documents get chunked, chunks get embedded, embeddings go into a vector database, and at query time you run a similarity search, grab the top-k chunks, and stuff them into the prompt. It works well enough to demo. It falls apart once you move past a handful of short documents and into real enterprise content: hundreds of pages, multiple documents, sections that reference each other, information buried on page 40 that your top-5
Where I go deep
Agentic Systems
Multi-agent orchestration, supervisor patterns, memory systems, tool routing. Designing autonomous loops that do real work - not demos.
LLM Infrastructure & Cost Governance
Semantic routing, SLM/LLM hybrid stacks, observability pipelines, token cost reduction. Making AI deployable at scale without the LLM Tax eating your margins.
AI Product Engineering
Spec-driven development, eval frameworks, RAG pipelines, production deployment patterns. The full system - not just the model layer.
Enterprise AI Adoption
Architecture reviews, build-vs-buy frameworks, AI governance, team structure. The decisions that determine whether an AI investment succeeds or stalls.
Tools of the trade
What I'm Thinking About
Working through what "memory" actually means for a long-running agent. Episodic? Semantic? Neither pattern from human cognition maps cleanly.
Enterprise teams are spending 4–6× what they should on inference. The answer isn't a cheaper model - it's a smarter router.
How do you debug an agent that's three hops deep in a tool-use loop? The tracing primitives don't exist yet.
Pick my brain
I do a limited number of 1:1 sessions - on AI architecture, building defensible AI products, technical strategy for non-tech founders, and career decisions in tech. 25 years of context, no slides, no fluff.
Fractional CTO & Tech Advisor
Expert tech review for non-tech founders. Save dev cost before you spend it.
Build an AI Moat, Not a Wrapper
Turn your thin wrapper into a defensible AI asset.
Career Mentorship
Career clarity through honest conversation. 25 years of pattern-matching.