Sushil Rajthala runs an insurer’s Kubernetes platform from Kathmandu: sixteen microservices, their pipelines, certificates, databases, monitoring and backups, with a security-first hand on every layer.
4+years on one production platforminsurance, since 2022
16microservices kept runningidentity to claims to finance
3runtimes in one estateKubernetes · compose · bare metal
24/7monitored and alertedPrometheus · Loki · Teams
How I run a platform
Four loops, every week
Each panel is a rendering of a real piece of the platform, with sample values.
Ship
Pipelines with gates, not hope
GitLab and Jenkins build the images, scan dependencies and layers, handle secrets outside the repo, and stop at a deployment gate before production.
pipeline release-2026.106 stages
buildpassed
scan imagepassed
scan depspassed
pushpassed
deploy uatrunning
deploy prodgated
Run
Workloads on Kubernetes and Docker
Deploy and troubleshoot the insurance services: workload configuration, scaling, pod and service diagnosis, ingress and DNS, TLS from Certbot.
kubectl get pods -n insurance16 pods
NAME
READY
STATUS
identity-api
1/1
Running
underwriting-api
1/1
Running
claims-api
1/1
Running
reinsurance-api
1/1
Running
accounting-api
1/1
Running
crm-api
1/1
Running
organization-api
1/1
Running
agent-api
0/1
Pending
finance-api
1/1
Running
Watch
Metrics, logs and alerts that wake the right person
Prometheus and Grafana for the numbers, Loki for the logs, Uptime Kuma and a log watcher for downtime and 500s, pushed to email and Microsoft Teams.
loki · tail {namespace="insurance"}live10:42:07INFO claims-api GET /claims/renewals 200 118ms10:42:09INFO identity-api POST /connect/token 200 41ms10:42:11WARN notification POST /sms 502 upstream timeout10:42:11WARN watcher: 3× 5xx on notification in 60s10:42:12INFO alert sent: Teams #platform-ops, mail ops@10:42:40INFO notification POST /sms 200 220ms10:43:12INFO watcher: notification recovered
Extend
New capability without new risk
A self-hosted 27B model on an NVIDIA L40S, served with vLLM in Docker, turns the company's operations workbook into an insights page for management.
nvidia-smi · vllm serveL40S
GPU
NVIDIA L40S, 45 GiB
model
Qwen3.6-27B, FP8
engine
vLLM, Docker, tp=1
context
131,072 tokens
features
prefix cache, tool parser
status
serving on :8000
Work
Twenty things built, run, or broken and fixed
Platforminsurance-k8s
Insurance microservices on Kubernetes
Deploys and troubleshoots identity, underwriting, claims, reinsurance, accounting and CRM services on Kubernetes and Docker: workload config, scaling, pod and service diagnosis, ingress availability.
Platformuat-stack
UAT stack with Nginx reverse proxy
A docker-compose UAT environment for the whole .NET insurance platform: compose overlays, environment files, Nginx routing per subdomain and Let's Encrypt certificates mounted in.
Securityci-gates
GitLab and Jenkins pipelines with scanning
Container image builds with dependency and image scanning, secrets handling and deployment gates, following release practices that reduce downtime.
Operationsobservability
Prometheus, Grafana and Loki
Metrics, dashboards, centralized logs and alerting across the cluster, with Rancher monitoring for the cluster itself.
Operationsalerting
Downtime and 500-error alerting
Lightweight uptime checks and a log watcher that raise email and Microsoft Teams alerts when a service goes down or starts returning 500s.
Securitytls
DNS, Ingress and TLS certificates
DNS resolution, Nginx and Ingress routing, Certbot-managed certificates and internal service access across Linux and Windows Server.
Datapostgres
PostgreSQL operations and reporting
Connection and session checks, JSON and JSONB querying, and SQL reports and dashboards that give leadership repeatable visibility into policy, claims and renewals.
Operationsbackup
Backup and recovery planning
Backup and recovery for Kubernetes workloads, databases, virtual machines and shared storage, including MinIO object storage.
AIllm-l40s
Self-hosted LLM on an NVIDIA L40S
Serves a 27B-parameter FP8 model with vLLM and SGLang in Docker on a shared 45 GB GPU, tuned around co-tenant memory limits and a vendor GPU-sharing shim.
AIinsights
Operations insights from Excel
The hosted model turns the company's finalized operations and profit-and-loss workbook into an insights page for management.
AIlocal-llm
Local models and agent harnesses
Runs small models locally through LM Studio and wires them into agent harness UIs as OpenAI-compatible providers.
Dataautomation
Automation scripts with validation
Python and SQL scripts with built-in validation checks that source data from PostgreSQL, APIs and application logs.
Securitybug-bounty
Web-application bug bounty
Recon, Burp Suite and Metasploit work on web targets, built on Certified Ethical Hacking training and lab practice on Kali Linux.
Securityleak-monitor
Credential-leak monitoring platform
A dark-web and credential-leak monitoring service, in design.
AIagent-platform
Agent platform for construction
Evaluating multi-model agent harnesses with an editable end-user UI for construction and real-estate teams.
Operationsnewsletter
Staff newsletter, HEI Insights
A recurring staff newsletter sent by email to the whole company.
Platformdatacentre
Tier III data-centre design
University coursework on data-centre infrastructure: network and Tier III design, hardware and space dimensioning, electrical, mechanical and DCIM.
Platformwindows
Windows Server administration
Windows Server 2019 administration alongside Linux, with secure configuration and access control.
Platformlinux
Linux system administration
Linux servers, networking and hardening, from college labs to production.
Operationsrelease
Release coordination with vendors
Coordinates releases, validates changes and resolves production issues with DevOps, IT and external vendors across the insurance apps and customer portals.
Console
Ask the terminal
It answers from the same facts as the rest of this page: what he runs, what he has built, how to reach him.
sushil@kathmandu: ~bash
Last login: a moment ago from your browsersushil@kathmandu:~$ helphelp list the commands
whoami who runs this console
kubectl get pods services on the platform
ls work the index of work
cat work/<slug> one piece of work
cat tools what is used for what
cat training courses and degrees
contact how to reach Sushil
uptime how long on the platform
clear clear the screensushil@kathmandu:~$
Toolkit
What runs on what
Run
Kubernetesworkloads, ingress
Dockerimages, compose
Rancherclusters
Linuxservers
Windows Server 2019domain, access
Ship
GitLab CIpipelines
Jenkinsbuilds
Pythonscripting
Watch
Prometheusmetrics
Grafanadashboards
Lokilogs
Uptime Kumachecks, alerts
Secure
Nginx, Ingressrouting
Certbot, TLScertificates
DNSresolution
Burp Suite, Kalitesting
Store
PostgreSQLSQL, JSONB
MongoDBdocuments
MinIOobjects
Serve models
vLLMFP8 serving
SGLangserving
LM Studiolocal models
About
Sushil Rajthala
I came to infrastructure through an insurance company that needed its platform to stay up. Since 2022 I have worked beside the DevOps and IT teams at Himalayan Everest Insurance: shipping releases to Kubernetes, keeping certificates and DNS honest, watching the cluster through Prometheus and Loki, and writing the SQL that tells leadership what the business did this month. Ethical-hacking training shapes how I configure things. I am looking for a DevSecOps, platform or infrastructure-security role where that mix is useful.
Based in
Lubhoo, Lalitpur
At HEI since
September 2022
Looking for
DevSecOps, platform, infra security
Training and certification
DevOps: Docker and Kubernetes, Udemy2025
Certified Ethical Hacking, Broadway Infosys2024
Windows Server 2019, Udemy2022
Certified Ethical Hacking, Udemy2020
Linux administration and networking2020
Python basics and intermediate2019
Education
BSc CSIT, Tribhuvan University2021
Sagarmatha College of Science and Technology2021
Milestone International College, Science2015
Mahalaxmi Academy, SLC, 88%2013
Contact
Need someone on the platform?
Open to DevSecOps, platform operations and infrastructure security roles, on site in Kathmandu or remote. Reachable Sunday to Friday, 09:00 to 18:00 NPT.