Building Fintech That Reconciles to the Paisa
Ledgers, idempotency and daily reconciliation. The unglamorous engineering that separates fintech products that survive audit from those that do not.
9 min read
50 articlesNotes from
Notes from
building things.
What we have learned shipping AI systems, SaaS platforms and industry software — including the parts that went wrong. Written by Purushottam Kumar Suman.
LatestEngineering
The Strangler Pattern in Practice: Replacing a System Without Stopping It
Big-bang rewrites fail for reasons that are structural, not managerial. How to route traffic away from a legacy system feature by feature, reversibly.
Purushottam Kumar Suman10 min read
Onboarding Is Where Trials Are Lost
Most SaaS churn happens before anyone uses the product properly. How to find the moment people stop, and design the path to first real value.
8 min read
What an MVP Is, and What It Definitely Is Not
Not a cheap version of the product and not a prototype. An MVP is an experiment with a hypothesis — and most of the value is in what you agree not to build.
8 min read
Why Most AI Pilots Never Reach Production
The demo impressed everyone and then the project quietly died. Here is the pattern behind it, and the four things that separate pilots that ship from pilots that do not.
8 min read
The Queries That Get Slow at Scale, and Why
N+1s, missing indexes, unbounded results and the ORM conveniences that quietly stop scaling. Diagnosis and fixes with the reasoning behind them.
9 min read
Healthcare Software Clinicians Will Actually Use
Clinical software fails on click count, not features. Designing for short consultations, constant interruption and staff who will route around friction.
8 min read
Empty States Are Not Edge Cases
Every user sees the empty state before they see the full one. Designing the zero-data, error and loading screens that decide whether people continue.
7 min read
RAG vs Fine-Tuning: Choosing Correctly the First Time
Teams reach for fine-tuning when they need retrieval, and for retrieval when they need neither. A practical decision framework based on what each technique actually changes.
9 min read
Hiring Your First Engineers Without a Technical Founder
How to assess technical skill you cannot evaluate directly, what to test for, and the hiring mistakes that are hardest to undo at small scale.
9 min read
Zero-Downtime Deploys That Actually Work
Rolling releases are the easy half. The hard half is database migrations, and that is where most 'zero-downtime' pipelines quietly fail.
9 min read
What Manufacturing Data Actually Tells You
Getting data off the plant floor is the easy part. Turning it into decisions about downtime, yield and maintenance is where the value is.
9 min read
Design for the Mid-Range Android Phone, Not Your Laptop
In India the median user is on a mid-range device with variable connectivity. What that means for images, fonts, JavaScript and interaction design.
8 min read
Building an LLM Evaluation Harness That Actually Catches Regressions
Without scored evaluation, prompt engineering is superstition. How to build a golden set, choose metrics that mean something, and wire it into CI.
10 min read
Build or Buy: A Framework You Can Defend
Building what you should buy wastes a quarter. Buying what you should build caps your ceiling. Four questions that settle it.
8 min read
What Server Components Actually Changed
Beyond the discourse: where the rendering boundary belongs, what it does to bundle size and data fetching, and the mistakes that follow from misplacing it.
9 min read
The ETA Problem: Honest Estimates Beat Confident Ones
Customers do not need a precise arrival time. They need one they can trust. Building ETAs that account for reality, and communicating uncertainty well.
8 min read
Accessibility Is Much Cheaper Before Launch
Semantics, focus order, contrast and keyboard paths cost little during the build and a great deal afterwards. A practical WCAG AA working method.
8 min read
Controlling LLM Costs Before They Control Your Roadmap
AI features that work but cannot be afforded get switched off. Caching, model tiering, context discipline and per-tenant budgets — with the numbers that matter.
8 min read
Scoping a Software Project So It Does Not Overrun
Overruns are usually scoping failures, not engineering ones. What to pin down before a quote, and the questions that surface hidden work.
9 min read
API Versioning Without Breaking Anyone
Additive change, deprecation policies you can enforce, and usage telemetry that tells you when it is genuinely safe to remove an endpoint.
8 min read
EdTech That Teachers Actually Adopt
Teachers are the most time-poor users in any system. Why tools that add preparation work fail regardless of quality, and what removing work looks like.
8 min read
Deciding What to Build Next, With Evidence
Roadmaps built from the loudest request produce products nobody loves. A method for weighing requests against usage, revenue and cost of delay.
8 min read
AI Agents: What Actually Works in Production
Autonomous agents make compelling demos and unnerving production systems. Where agents genuinely earn their place, and the boundaries that make them safe.
9 min read
Working With an Agency Without Losing Control
How to keep ownership of your product, code and knowledge while an external team builds it — from a company that takes this work.
8 min read
Idempotency: The Habit That Prevents Double Charges
Networks retry, users double-click, queues redeliver. The single most valuable defensive pattern for any endpoint that moves money or creates records.
7 min read
One Stock Number: Inventory Truth Across Channels
Overselling, dead capital and cancelled orders all trace to the same cause. Building a single authoritative stock position that every channel respects.
8 min read
Pricing Pages That Do Not Confuse People
Confusion converts worse than a high price. Structuring plans, choosing the value metric, and removing the questions that stall a decision.
7 min read
Grounding: The Practical Cure for Confident Wrong Answers
Hallucination is not a model defect to wait out. It is a systems problem with known mitigations — retrieval, structure, refusal design and citation.
8 min read
In-House or Outsourced: Choosing by Time Horizon
The question is not which is better but which fits the shape of the work. Duration, specialisation and the cost of a wrong hire.
7 min read
Observability Beyond Dashboards Nobody Opens
Alert on symptoms users feel, trace the requests that matter, and log for the question you will ask at 2am — not for completeness.
8 min read
Where Claims Automation Should Stop
Straight-through processing for the simple majority, humans for the rest. Drawing the boundary so speed does not become an unmanaged liability.
8 min read
Design Systems That Survive Three Teams and Two Years
Most design systems decay into a component library nobody follows. What makes tokens, documentation and governance actually hold as a product grows.
8 min read
Frontier or Open-Weight? A Cost and Control Decision
The choice is rarely about raw capability. It is about cost at your volume, data residency, latency and who is responsible when the endpoint is down.
8 min read
How Runway Should Change Your Engineering Decisions
The right architecture at eighteen months of runway is the wrong one at five. Matching technical decisions to the time you actually have.
8 min read
Test Suites People Actually Trust
A flaky suite is worse than no suite, because it teaches the team to ignore red. How to build coverage that earns confidence and keeps it.
8 min read
Agritech for 2G Networks and Shared Phones
Most agritech fails on assumptions about connectivity, literacy and device quality. Building for the conditions farmers actually work in.
8 min read
The First Run Decides Retention More Than Features
Day-one experience predicts day-thirty retention better than functionality does. What to fix first when people try the product once and never return.
7 min read
Prompts Are Code. Treat Them That Way.
Version control, review, tests and staged rollout for prompts. The engineering hygiene that turns prompt tinkering into a repeatable practice.
7 min read
Choosing a Technical Co-founder, or Deciding Not To
Equity is the most expensive currency a founder has. When a technical co-founder is right, and when a partner or a hire serves you better.
8 min read
Caching Decisions That Survive Real Traffic
What to cache, where to put it, and how to invalidate without serving stale data — plus the stampede that takes systems down at exactly the wrong moment.
8 min read
Construction Apps for Gloved Hands and No Signal
Site software fails on ergonomics before it fails on features. Designing capture that takes seconds in sunlight, with gloves, offline.
7 min read
Microcopy That Removes Hesitation
Button labels, field hints, confirmations and error messages. The smallest text in the product does the most work at the moments that matter.
7 min read
Document Intelligence: Getting Structure Out of Messy PDFs
Invoices, statements and forms in every layout imaginable. A practical pipeline for extraction with confidence scoring and a human path for the rest.
9 min read
Why Fixed-Price Projects Go Wrong for Both Sides
Fixed price feels safe and creates incentives that damage the product. When it works, when it does not, and the middle ground that usually does.
8 min read
Technical Debt Is a Budget, Not a Sin
Some debt is a rational trade and some is compounding interest. How to tell them apart, and how to argue for repayment in terms a business will fund.
8 min read
Public Sector Software Has to Work for Everyone
Citizen services cannot choose their users. Designing for old devices, low literacy, multiple languages and people with no other option.
8 min read
Prototype Before You Commit Engineering
A week of prototyping routinely saves a quarter of building. What to prototype, at what fidelity, and how to test it so the answer is trustworthy.
7 min read
When Not to Use AI
An AI company arguing against AI. The situations where a rule, a query or a form beats a model — and how to tell before you have spent the budget.
7 min read
Your First Hundred Customers Should Be Manual
Automating before you understand the workflow builds the wrong thing efficiently. Why doing it by hand first is an engineering strategy, not a stopgap.
7 min read
Reading is easier than building.
If one of these describes a problem you have,
bring it to a call and we will talk it through.
AI-First Engineering
Secure & Scalable
Built to Deliver Impact