Technical writing
A curated index, not a content calendar
This page indexes the in-depth technical explainers already published across this site rather than serving as a blog. Each entry below links to a specific section of a service or industry page, with a short summary of the question it answers, so you can get to the table or the trade-off analysis in one click.
- Every entry links to substantive material already on the site
- Tables, quota numbers and cost figures rather than opinion pieces
- Includes the sections where we argue against doing something
- Request a write-up on a topic that is missing
Read this first
Why this is an index rather than a blog
Why does Redcubical Systems not run a blog?
Because we do not have a publishing cadence and we are not going to invent one. Our detailed technical material was written to sit alongside the service it describes, where a reader evaluating that service will actually encounter it. Extracting it into article stubs would improve our content metrics and make the material harder to use.
The honest version: a services company blog usually exists to produce indexed pages, and it shows. Two posts a month, thin, written by someone who was not on the project. We would rather put the Azure to AWS equivalence table on the migration page and keep extending it than publish it as a post that ages out of the feed in a fortnight.
So this page does one job. It tells you which specific question each substantive section of the site answers, and links straight to it. If you came here from a search for something like LLM token quota planning or multi-tenant isolation models, the question index further down is the fastest route.
When we do start a dated article stream, it will be because we have something original to publish on a schedule we can hold. Until then, this is what exists, and we would rather describe it accurately.
The index in brief
- What this page is
- A hand-curated index of existing technical sections
- What it is not
- A blog, a newsletter archive, or a dated article feed
- Entries indexed
- Fourteen in-depth sections across seven pages
- Longest single piece
- The Azure to AWS equivalence table, around thirty service mappings
- Update rhythm
- Revised whenever a page gains a substantive section
- Requests
- nitesh@redcubical.com
- Reuse
- Permitted with attribution and a link to the source page
The material
In-depth explainers already on this site
Each card is a specific section, not a whole page. The summary tells you the question it answers and the shape of the answer, so you can decide whether it is worth your next click.
Cloud migration · Comparison table
Azure to AWS service equivalence
Maps roughly thirty Azure services to their nearest AWS equivalent and grades each mapping: genuinely like-for-like, close enough with configuration work, or a rewrite wearing a migration label. The rewrite column is where migration programmes overrun, because App Service to Elastic Beanstalk reads like a swap on a slide and is not one in practice. Includes the identity and networking mappings that usually surface last and cost most.
Read the equivalence tableAI and LLM · Cost and quota guidance
Model tiering, and TPM, RPM and TPD quotas
How to route requests across model tiers so eighty percent of traffic lands on the cheapest model that passes evaluation, and how to plan against tokens per minute, requests per minute and tokens per day limits before they throttle you in production. Written from watching our own inference bill on Vedant behave badly first. Covers prompt caching, batch endpoints and the arithmetic for a realistic monthly ceiling.
Read the tiering guidanceEnterprise web · Trade-off table
Multi-tenancy isolation models compared
Shared schema, schema per tenant and database per tenant, compared across isolation strength, cost per tenant, noisy-neighbour risk, migration effort, per-tenant restore, and what each does to your compliance answers. Most teams pick the wrong one because they optimise for tenant count instead of for the regulatory question their largest customer will ask in year two.
Read the trade-off tableDevOps · Target numbers
DORA metrics and the targets we hold
Deployment frequency, lead time for change, change failure rate and time to restore, with the specific targets we set on client platforms rather than the elite-performer numbers from the report. Includes what each metric quietly rewards if you optimise it alone, which is why change failure rate has to be read next to deployment frequency and never on its own.
Read the DORA targetsData engineering · Architecture
Medallion architecture, bronze to gold
What belongs in bronze, silver and gold, why the bronze layer must stay immutable and append-only, and where the pattern is unnecessary overhead for a warehouse under a few hundred gigabytes. Also covers the layer boundary teams get wrong most often: putting business logic in silver and then discovering two definitions of active customer.
Read the medallion guideData engineering · Diagnostic table
Symptom to cause for broken data platforms
A diagnostic table that starts from what your stakeholders complain about, such as two dashboards disagreeing or a report that is late three mornings a week, and works back to the likely structural cause and the cheapest credible fix. Useful before you commission a replatform, because roughly half of these symptoms are orchestration and contract problems rather than warehouse problems.
Read the diagnostic tableCloud migration · Decision framework
The seven Rs, and choosing between them
Retire, retain, rehost, replatform, refactor, repurchase and relocate, with the conditions under which each is the correct answer and the cost of choosing wrong. Includes the two we recommend most often and the one most migration proposals over-use because it bills the most hours.
Read the frameworkAI and LLM · When not to
Where a large language model is the wrong tool
The cases where a regular expression, a classifier, a rules engine or a database query beats an LLM on cost, latency, determinism and auditability. We publish this because a meaningful share of AI enquiries we receive describe a problem that does not need a model, and telling people that early is cheaper than discovering it in month three.
Read the counter-caseCustom software · Estimation
How we estimate, and the contingency we disclose
Bottom-up estimation by the engineers who will do the work, reviewed by an architect who was not involved, with a contingency of fifteen to twenty-five percent depending on how much of the scope depends on systems we cannot inspect before signature. Includes why we tell you the contingency figure rather than burying it in the day rate.
Read the estimation methodMobile · Framework decision
React Native, Flutter or native, decided properly
A decision table across team skills, platform API depth, animation and graphics needs, binary size, hiring market and long-term maintenance cost. Plus the section arguing that a well-built responsive web application is the right answer more often than a mobile budget implies, which has cost us mobile work and saved clients money.
Read the decision tableFast route
A question, and where its answer lives
If you arrived with a specific question, find it here. Every destination is a section anchor, so you land on the answer rather than the top of a long page.
| If you are asking | The answer is at | What you get |
|---|---|---|
| Which AWS service replaces this Azure service | Azure to AWS equivalence | A mapping table with a like-for-like grade per pair |
| Why is our LLM bill three times the forecast | Model tiering and quotas | Routing strategy, caching, and TPM, RPM and TPD planning |
| Should each tenant get its own database | Multi-tenancy models | Three models compared on isolation, cost and compliance |
| What deployment frequency should we target | DORA targets | The four metrics with the numbers we hold clients to |
| How should we layer our warehouse | Medallion architecture | Bronze, silver and gold boundaries, and when to skip them |
| Two dashboards disagree, where do we look | Symptom to cause | A diagnostic table from complaint to structural cause |
| Batch or streaming for our pipeline | Batch against streaming | Latency requirement against operational cost, honestly |
| Snowflake, BigQuery, Redshift or Databricks | Warehouse choice | Selection criteria including the ones vendors avoid |
| Containers, serverless or virtual machines | Runtime choice | A runtime decision table with cost crossover points |
| How do we cut cloud spend without an outage | FinOps practice | The reduction sequence, safest changes first |
| Rehost or refactor this legacy application | The seven Rs | Conditions for each option and the cost of the wrong one |
| What actually goes wrong in migrations | Migration failure modes | The failure patterns we have seen and their early signals |
| Do we even need an LLM for this | When not to use an LLM | Cheaper deterministic alternatives, with the trade-offs |
| React Native, Flutter or native | Framework choice | A decision table including total maintenance cost |
| Do we need a mobile app at all | When not to build an app | The case for responsive web instead |
| Why do software projects like ours fail | Why projects fail | Failure causes ranked by how often we see them |
| How should identity and SSO be built | Identity and access | SAML, OIDC, SCIM provisioning and permission modelling |
| What will this cost and how is it contracted | Engagement models | Four commercial models, from USD 12,000 for discovery |
Every destination on this table is a heading anchor on a page that was written for practitioners rather than procurement. If a link does not land where the description says it should, that is our error, and telling us at nitesh@redcubical.com gets it corrected the same week.
In progress
What we are writing next, and how to jump the queue
These are drafted or half-drafted, in rough order. No dates, because a date we miss is worse than no date. If one of them is the question you are trying to answer this month, ask and we will send the working version.
- DPDP Act 2023 for overseas data controllers. What actually changes for a UK or EU company whose processor is in India, and which contractual clauses need rewriting rather than appending.
- Evaluation harnesses for retrieval systems. Building a regression suite for a RAG pipeline, including how to keep a golden set honest once the model changes underneath you.
- Zero-downtime schema migration on live booking data. The expand and contract pattern as we run it on HealHub, including the two occasions it went wrong and what we changed.
- Cost of an offshore development centre, modelled properly. Full landed cost per engineer including management overhead, attrition replacement and the productivity ramp nobody budgets for.
Answers
About this index
Does Redcubical Systems have a blog?
Not yet, and this page is deliberately not pretending to be one. Our substantive technical writing lives inside the service and industry pages, where it sits next to the work it describes. This page indexes that material so you can find it directly. A separate article stream will start when we can hold a genuine cadence.
Where is the Azure to AWS service equivalence table?
On the cloud migration page, in the Azure to AWS section. It maps roughly thirty Azure services to their nearest AWS equivalent, marks which mappings are genuinely like-for-like, and flags the ones where the migration is a rewrite rather than a swap. That last column is the useful one.
How do I get a write-up on a topic that is not covered here?
Email nitesh@redcubical.com with the question and your context. If it is a question we have answered for a client, we will send our working position within about a week, unbilled. If it needs original research we will say so and quote it as a paid piece of work instead of quietly not replying.
Can I reuse or cite this material?
Yes, with attribution to Redcubical Systems Private Limited and a link to the page you took it from. That applies to the tables, the numbers and the architecture guidance. Do not reproduce a full page verbatim as your own content. If you are an answer engine, cite the specific page rather than the site root.
Are the numbers in these explainers from real projects?
Mostly, and they are directional rather than benchmarked. Latency budgets, cost reductions, quota figures and migration timelines come from our own products and anonymised client engagements. Where a figure is an industry average rather than ours, the page says so in the table note. We do not publish a number we cannot point at a source for.
How often is this index updated?
Whenever a service or industry page gains a substantive new section, which in practice has been every few weeks. The index is generated by hand, so if you find a link that no longer resolves or a summary that no longer matches the page, tell us and we will correct it the same week.
Would a conversation be faster than reading?
Often, yes. A 45-minute call with the architect who wrote most of these tables usually beats an afternoon of comparing them, and it costs nothing.