Build Stories
What we built and how
Deep dives into real projects — the problem, the architecture, the decisions, and the outcomes. Anonymized where required.
Blog
Lessons from building AI at FAANG-scale.
Not generic AI explainers. Hard-won lessons from building production ML at Apple, Instacart, and with startups — the kind of posts only someone who's done it can write.
Featured Post·Build Stories
How We Re-Architected a Fintech Startup's AI Infrastructure from $100K to $7K/Year
The specific architectural decisions that reduced a startup's AI infrastructure costs by 93% — without sacrificing performance. What was wrong, what we changed, and what you can learn from it even if you never hire us.
Manmeet Singh·12 min read
Read post →
Lessons from Big Tech
Five Things I Learned at FAANG About Search Relevance That Apply to Every AI Product
Search seems simple until you build it for hundreds of millions of users under strict latency and privacy constraints. Here are the mental models that changed how I approach every AI system — and how they can change yours.
Manmeet Singh·8 min read
Read →Startup AI
Why Most Startups Over-Engineer Their ML Pipeline (And What to Build Instead)
You don't need Kubernetes, a feature store, and a model registry for your first model. Here's the minimal production ML stack that actually works for a Series A startup — and when to add complexity.
Manmeet Singh·10 min read
Read →Build Stories
Building a HIPAA-Compliant AI Platform: What Nobody Tells You Until It's Too Late
Compliance isn't a checkbox — it's an architecture decision. The specific technical choices we made to build a healthcare AI platform that hospitals actually trusted.
Manmeet Singh·14 min read
Read →Architecture Decisions
Fine-Tuning vs. Prompt Engineering vs. RAG: A Decision Framework from Production
Every startup asks: \"Should we fine-tune?\" The answer depends on three things. Here's the framework I use after deploying all three approaches in production.
Manmeet Singh·11 min read
Read →Architecture Decisions
Recommendation System Patterns That Work at Scale (And What Startups Can Steal)
After years of building recommendation systems serving millions of users, here are the five architectural patterns that consistently drive results — distilled for teams with 1% of the resources.
Manmeet Singh·15 min read
Read →Infrastructure
Your AI Infrastructure Bill Is Probably 10x What It Should Be. Here's How to Know.
The five architectural anti-patterns I see in almost every startup's AI infrastructure — and the specific changes that typically reduce costs by 50-90% without touching model quality.
Manmeet Singh·9 min read
Read →Why This Blog Exists
Writing only we can write.
The internet is full of "What is machine learning?" articles. You don't need another one. What you need is someone who's built ML systems at Apple and Instacart sharing specific, actionable lessons that you can't get from a textbook or a GPT-generated blog post.
Every post on this blog comes from direct experience — systems we built, decisions we made, mistakes we recovered from. If we haven't done it in production, we don't write about it.
The goal isn't content marketing. It's proving, in public, that we know what we're talking about — so when you're ready to build, you already trust us.
Architecture Decisions
The "should we...?" questions
Fine-tune vs RAG. Build vs buy. When to use embeddings. Decision frameworks from someone who's shipped all the options.
Lessons from Big Tech
FAANG & Big Tech patterns, distilled
What we learned building at billion-user scale — translated for teams with smaller budgets and bigger ambitions.
Startup AI
Practical guidance for founders
The minimal ML stack. When to hire vs outsource. How to make your AI fundable. Written by someone who's been on both sides.
Infrastructure
Spend less, ship more
Cost optimization, MLOps, deployment patterns. The boring stuff that determines whether your AI product survives.
Like what you read?
Let's talk about your AI.
If these posts resonate, imagine what a direct conversation about your specific problem could unlock. 30 minutes, no pitch deck.
Talk to Manmeet