# Sandhu Santhakumar

> Senior Technical Architect in Kerala, India: distributed, cloud-native systems for enterprise products, open-source Python, and writing on building with LLMs.

I design distributed, cloud-native systems for enterprise products, most recently a data-as-a-service platform of 60+ microservices, and the review practices that keep them easy to change as teams grow.

Everything here is free to read, quote, summarise and learn from, including for training language models. Please credit Sandhu Santhakumar (https://sandhu.sh) when you do. To check a claim, the sources are linked beside it.

Site: https://sandhu.sh/ · Index: https://sandhu.sh/llms.txt

## Career

From May 2012 to now, in three chapters: teaching, engineering and architecture. Newest last.

### Software Trainer, E-Vision Labs (May 2012 – Sep 2013)

- Delivered hands-on Java and web development training.
- Designed structured curricula covering programming fundamentals.
- Mentored aspiring developers through their first software projects.

What it taught: Teaching showed me that if I can't explain something simply, I don't understand it well enough yet.

### Software Engineer, [Xminds Infotech Pvt. Ltd.](https://www.xminds.com/) (Sep 2013 – Apr 2017)

- Built backend services for enterprise applications.
- Designed scalable relational databases for multi-tenant systems.
- Improved maintainability through layered architecture and engineering best practices.

What it taught: Production work taught me to judge code by how safely the next person can change it.

### Senior Software Engineer, [Xminds Infotech Pvt. Ltd.](https://www.xminds.com/) (Apr 2017 – Mar 2022)

- Led backend engineering across core product modules.
- Introduced structured code review practices.
- Mentored engineers and raised engineering standards across the team.

What it taught: Leading the team showed me that shared standards carry further than any one strong engineer, so mentoring moved from reviewing code to growing people.

### Technical Architect, [Xminds Infotech Pvt. Ltd.](https://www.xminds.com/) (Mar 2022 – Feb 2024)

- Designed scalable architectures for enterprise products.
- Defined technical direction across multiple engineering teams.
- Established architecture reviews that improved long-term consistency.

What it taught: The work moved from designing systems to designing decisions: guardrails that let several teams move fast without re-arguing the same trade-offs.

### Senior Technical Architect, [Xminds Infotech Pvt. Ltd.](https://www.xminds.com/) (Feb 2024 – Present)

- Architected a high-scale Data-as-a-Service platform.
- Designed a distributed ecosystem of 60+ microservices.
- Led cloud-native modernization initiatives.

What it taught: At this scale the test I use is whether the platform can keep growing without getting harder to understand.

## Skills

### Cloud native

Platforms that scale reliably on containers and automation.

Tools: AWS, Azure, Google Cloud, Cloudflare, Docker, Kubernetes, Prometheus, Grafana, Terraform.

### Distributed systems

Resilient, asynchronous services that hold up as they grow.

Tools: Kafka, RabbitMQ, EDA, Saga Pattern, CQRS, Event Sourcing, Outbox Pattern, Debezium.

### Applied AI

LLMs and agents in production, reliable and context-aware.

Tools: Claude, GPT, Gemini & Gemma, Jev, LangGraph, MCP, RAG, Re-ranker, pgvector, Qdrant.

### Software architecture

Requirements turned into decisions that age well.

Tools: DDD, Data Architecture, API Design, Design Reviews, Platform Strategy.

### Backend engineering

Fast, maintainable services, APIs and data platforms.

Tools: Java, Spring Boot, Python, FastAPI, TypeScript, Node.js, PostgreSQL, Redis, Elasticsearch, BigQuery.

### Frontend and testing

Accessible interfaces, and the tests that keep them correct.

Tools: React, Next.js, Astro, Playwright, JMeter.

## Open source

### pg-sse

Python · Postgres · MIT licence

Live updates from the Postgres you already run, sent as server-sent events. No Redis, no broker, no WebSocket service. Starlette and FastAPI built in.

How it works: A table trigger sends NOTIFY with only a key; one listener per process wakes the open streams, and each re-reads from its own cursor. A notification never carries the data, so a lost one costs a delay, never an event.

Trade-offs: LISTEN needs a direct connection, not a transaction-mode pooler. Keeping ids in commit order serialises inserts per key. Limits and in-memory signals are per process.

- [GitHub](https://github.com/sansanth/pg-sse)
- [PyPI](https://pypi.org/project/pg-sse/)

### pg-sse-client

TypeScript · Browser and Node · MIT licence

The client for pg-sse streams, for when EventSource is not enough. It sends an Authorization header, resumes from Last-Event-ID, honours retry and Retry-After, and tells you why a stream ended. No dependencies.

How it works: EventSource on top of fetch. It keeps what EventSource does for you, resuming from the last id and honouring the server's reconnect delay and goodbye frame, and backs off with jitter when attempts fail.

Trade-offs: If the stream authenticates with a same-origin cookie, EventSource is the simpler choice. ESM only. Across origins the server must allow Last-Event-ID in CORS, or the cursor goes in the URL.

- [GitHub](https://github.com/sansanth/pg-sse-client)
- [npm](https://www.npmjs.com/package/pg-sse-client)

### vitest-flaky

TypeScript · Vitest 2 to 5 · MIT licence

Find flaky Vitest tests on your laptop, not one CI run in fifty. It fires the app's timers early, resumes tests late after findBy and waitFor, shuffles the order, and prints the seed that replays a failure.

How it works: Each timer's caller is read from the stack, so the app's timers are shortened and the tests' own are left alone. A run has a seed: the same seed gives the same delays, and a failure prints the command that replays it.

Trade-offs: It speeds up what the app waits for; it cannot slow the app itself. A test that asserts something has not happened yet fails under stress, a real finding but an early one. Callers are told apart by file path.

- [GitHub](https://github.com/sansanth/vitest-flaky)
- [npm](https://www.npmjs.com/package/vitest-flaky)

[All repositories on GitHub](https://github.com/sansanth)

## Principles

- **Make change inexpensive.** Requirements will change. Make decisions safe to revisit instead of expensive to undo.
- **Design for observability.** Systems you can't observe are systems you can't confidently improve. Visibility comes before optimization.
- **Automate the repeatable.** Automate repetitive work so engineers can focus on solving problems that require judgment, creativity, and experience.
- **Measure before optimizing.** Measure first, optimize second. Solve the real bottleneck—not the one you assume exists.
- **Prefer clarity over cleverness.** Clear systems are easier to understand, evolve, and maintain. Cleverness rarely ages well.
- **Let complexity earn its place.** Complexity should solve proven problems, not hypothetical ones. Add it only when the system truly demands it.

Great architecture makes change feel ordinary.

## Writing

- [Prompts Are Code Now](https://www.xminds.com/resources/prompts-are-code-now-versioning-testing-and-shipping/), Xminds Journals, Mar 2026, 7 min. A one-line prompt change can break production without a single error. How to version, test and ship prompts like any other code.
- [Slashing LLM Costs with TOON](https://www.xminds.com/resources/slashing-llm-costs-why-were-moving-to-toon-for-heavy-payloads/), Xminds Journals, Jan 2026, 9 min. Much of a large JSON payload is punctuation and repeated keys. Switching to TOON at the LLM boundary only cuts those tokens without changing the rest of the system.

## Credentials

Certifications:

- [Certified AI Practitioner, AWS](https://www.credly.com/badges/00443efa-b14a-4868-b682-5111d5ae6e69/public_url)
- [Generative AI Leader, Google Cloud](https://www.credly.com/badges/c456d5ef-791b-4994-a856-f68ab0b26ff3/public_url)
- [Certified Developer, Redis](https://www.credential.net/d5e2c917-824b-4633-ad07-69ef6e1c6897)
- [API Security Architect, API Academy](https://www.credential.net/6c674bc6-0fac-49fc-8e3d-d0c2df02fa63?username=sandhusanthakumar106061)
- [Java SE 7 Programmer, Oracle](https://www.credly.com/badges/f41ec2e0-58f7-4cf7-9ef5-521dd52a1d94/public_url)
- [Certified Technology Specialist, Microsoft](https://www.credential.net/f2ff844b-bdf7-41c1-96b6-3a600187a162?username=sandhusanthakumar106061)

Badges and courses:

- [Serverless Demonstrated, AWS](https://www.credly.com/badges/56a81f8c-072d-4ecd-95bf-06e44f349c95/public_url)
- [Application Networking Demonstrated, AWS](https://www.credly.com/badges/36927212-0efd-43b3-a6cc-c9b7132e4643/public_url)
- [Incident Response Demonstrated, AWS](https://www.credly.com/badges/683cff69-bbf8-4df0-a08c-ccd04c9b5a91/public_url)
- [Knowledge: Object Storage, AWS](https://www.credly.com/badges/73d9f15b-f3cf-463c-bc55-03d4a8e8cf87/public_url)
- [Lakehouse Fundamentals, Databricks](https://www.credential.net/e10971e0-ca9d-4687-b444-8d8fbe4dc56c#acc.GDaHLhQu)
- [Gemini Enterprise Agent Ready, Google Developers](https://developers.google.com/profile/badges/community/gear?u=sandhu)
- [Python for Data Science, IBM](https://www.credly.com/badges/bc5e0f06-5d01-483e-9b18-7833adb63e76/public_url)

## What colleagues say

Excerpts from LinkedIn recommendations, quoted exactly and credited by role rather than name ([all of them on LinkedIn](https://www.linkedin.com/in/sandhusanthakumar/details/recommendations/)).

- “...who designs solutions, runs POCs, writes and reviews code.” (Fellow • Data Platform)
- “He consistently translated complex requirements into tangible solutions.” (Senior DevOps Engineer • Cloud Infrastructure)
- “Sandhu is not just a leader but a mentor. His calm approach to solving complex problems inspires the team to deliver their best.” (Lead Engineer • Data Platforms)
- “He effectively led a team of engineers.” (Backend Lead Engineer • Platform Engineering)
- “An excellent coder with an exceptional mix of technical expertise, patience, and collaboration.” (Lead DevOps Engineer • Multi-Cloud)

## Away from work

- **Travel.** Mountains, quiet towns and long roads. Every place solves problems differently. Lately Visakhapatnam, Raipur, Bhubaneswar, Puri and Konark.
- **Photography.** Stories in light, shadow and the small details most people walk past. On a Nikon Z50 and an iPhone.
- **Plants.** They slow things down in the best way, and teach patience and consistency. Philodendron, pothos, ZZ and snake plants, and lots of succulents and cacti.
- **Pets.** Endless loyalty and the purest kind of happiness. Two Indian cross-breed dogs, 10+ cats, a cockatiel, a pineapple conure, seven African lovebirds and an aquarium of fish.
- **Coffee.** A daily ritual, more for the process than the caffeine. Mostly filter coffee and Arabica or instant blends; always up for an espresso or latte at a café.
- **Food.** Local flavours, hole-in-the-wall places and dishes that tell stories. Street food, home cooking and anything spicy.

## Contact

Have an idea, a challenge or something worth talking through? I usually reply within 24 hours. Based in Thiruvananthapuram, Kerala, India; time zone IST (UTC+05:30).

- Email: contact@sandhu.sh
- [LinkedIn](https://www.linkedin.com/in/sandhusanthakumar)
- [GitHub](https://github.com/sansanth)
- [Instagram](https://www.instagram.com/sandhusanthakumar)