/about · the long version

about .

~/about ● 7 roles ● 5 milestones
Dipankar Das

fig.02 / portrait, b&w 2025

/intro

Dipankar Das,
platform engineer who designs systems people can depend on.

Building sustainable and efficient platforms without firefighting. I work at the boundary between application code and infrastructure, with a focus on developer platforms, CI/CD, Kubernetes, observability, and system design.

Most of my work is about reducing cognitive load for the engineers who depend on a platform: removing host-level surprises, making deployments boring, designing systems that fail predictably, and giving teams enough telemetry to understand a problem before they reach for a workaround.

/education

B.Tech Computer Science Class of 2024
/work-history

Roles, in reverse chronology

01 / 07 Jul 2026 - Present
Green Software Foundation

Green Software Foundation ↗

Assembly Participant - OpenTelemetry SCI

  1. .01

    SCI for OpenTelemetry Semantic Conventions

    why
    The OpenTelemetry and green software communities needed a shared, consensus-based approach for representing the Software Carbon Intensity (SCI) specification, an ISO standard, in OpenTelemetry semantic conventions.
    what
    Represented ksctl/Kubmin as one of 40 practitioners, maintainers, and researchers contributing to the Green Software Foundation's SCI for OpenTelemetry blueprint. The Assembly is defining the design decisions needed to turn SCI into draft semantic conventions for review by the OpenTelemetry Semantic Conventions SIG. Contributed to discussions that reached consensus on Topic 1: Scope and Use Cases, including the "Improve, Then Compare, Then Adapt" approach, which prioritizes optimizing workloads first, followed by cross-provider comparison and real-time workload placement.
    how
    Participated in the Assembly's Discover, Deliberate, and Decide process, contributing a platform-builder perspective to the proposed standard for software carbon telemetry. This work is related to Kubmin's infrastructure efficiency and sustainability capabilities for Kubernetes workloads; Topics 2 and 3 are in progress.
02 / 07 Jun 2025 - Present
Kubmin - Ksctl

Kubmin - Ksctl ↗

Founder & Principal Engineer

  1. .01

    Research on Software Sustainability Metrics

    why
    Kubmin's workload intelligence required defensible choices about data sources, aggregation windows, formulas, algorithms, and the provenance of sustainability measurements.
    what
    Researched the scope, calculation methods, configuration model, and provenance requirements for Software Carbon Intensity, embodied carbon, and carbon intensity measurements.
    how
    Evaluated the underlying assumptions and measurement pipeline so Kubmin's sustainability scores and recommendations can be explained, reproduced, and connected to their source data.
  2. .02

    Distributed Orchestration Engine

    why
    Managing Kubernetes cluster lifecycles across AWS and Azure required a reliable, cloud-agnostic solution without heavy infrastructure overhead.
    what
    Architected a Go-based cloud-agnostic provisioning engine for full Kubernetes lifecycle management.
    how
    Built a high-availability state layer with Turso (Edge SQLite) and Redis, enabling idempotent cluster operations with a minimal infrastructure footprint.
  3. .03

    Event-Driven Task Orchestration with NATS

    why
    Long-running infrastructure tasks needed guaranteed execution without introducing the operational burden of heavy frameworks like Temporal.
    what
    Developed a lightweight event-driven state machine using NATS JetStream.
    how
    Implemented custom NAK/ACK and retry logic ensuring 100% task reliability during long-running cluster operations while keeping the system operationally simple.
  4. .04

    Relationship-Based Access Control (ReBAC)

    why
    Flat RBAC couldn't model the complex sharing hierarchies needed for multi-tenant teams across shared infrastructure.
    what
    Implemented a full Relationship-Based Access Control system using Authzed (SpiceDB).
    how
    Designed a hierarchical model spanning Org, Cluster, and Workload levels, enabling fine-grained permission enforcement and quota management across distributed engineering teams.
  5. .05

    Workload Intelligence & Cost Tracking

    why
    Teams had no per-workload visibility into what each deploy actually costs in money, energy, and compute — waste compounded silently with every release.
    what
    Built a per-workload cost and efficiency tracking engine powered by Kepler (CNCF) with SCI/SEE sustainability scoring aligned to the Green Software Foundation. Includes per-image-version deep profiling (hourly SCI, energy, CPU, memory, cost per image tag) and release-over-release efficiency diff.
    how
    Surfaces idle workloads, overprovisioned containers, and efficiency regressions between deployment versions — with dollar amounts and ready-to-use kubectl commands to fix them. Classifies each workload as CPU-bound, memory-bound, general, or resource-starved from runtime behavior. Compares workload costs across regions and instance types.
  6. .06

    Durable Workload Intelligence & Carbon Accounting

    why
    Workload cost, energy, and carbon metrics needed to remain queryable and comparable after the active observation window ended. Cache-based accumulation was not sufficient for reliable historical analysis.
    what
    Redesigned Kubmin's workload analysis pipeline around durable hourly records and snapshot finalization, adding operational and embodied carbon completeness tracking, provenance, retention controls, and image-level workload analysis.
    how
    Built a database-backed snapshot workflow using NATS and scheduled finalization, migrated analysis to relational records, added transactional updates and retry handling, and extended the frontend with workload detail, energy insight, and container/image views.
  7. .07

    AI-Agents Orchestration

    why
    Platform delivery velocity needed to scale without proportionally growing the team.
    what
    Orchestrated AI agents (Claude Code, Gemini CLI) into the core development workflow to autonomously handle well-scoped engineering tasks.
    how
    Leveraged agentic patterns — task decomposition, tool use, and iterative feedback loops — to boost team productivity by 40% while maintaining high code quality and architectural consistency.
  8. .08

    Guided Observability

    why
    Teams had monitoring stacks but no automated way to wire telemetry (metrics, logs, traces, energy) into workload cost and efficiency decisions. The gap between existing observability and actionable cost intelligence was manual glue work.
    what
    Designed an OpenTelemetry-native Guided Observability system for Kubmin — deploy missing backends or connect to existing Prometheus, Tempo, Loki, and Grafana stacks. Configures an in-cluster OTel Collector bridge for OTLP gRPC/HTTP alongside Kepler for energy attribution.
    how
    Built a setup workflow during cluster create/import that always installs the Kubmin agent, Kepler, Gateway API, and OTel Collector — while optionally deploying or connecting to existing monitoring backends. Wires telemetry into Kubmin's cost and efficiency ledger automatically, so metrics, logs, traces, and energy data support the same workload decisions.
  9. .09

    Existing Cluster Import

    why
    Teams running existing Kubernetes clusters (EKS, AKS, GKE, self-managed) couldn't add Kubmin's workload intelligence without migrating to a new cluster.
    what
    Built cluster import support for EKS, AKS, GKE Standard, and kubeconfig-based self-managed clusters — installing Gateway API CRDs, Kubmin agent, Kepler, and OTel Collector during the automated import workflow.
    how
    Implemented cloud-provider discovery for managed clusters (by cloud and region) and kubeconfig-based auth for self-managed clusters. The import workflow installs all required components programmatically, preserving the existing infrastructure ownership while adding workload intelligence on top.
  10. .10

    Workload Recommendation Engine

    why
    Teams had no structured way to understand the true profile of their running workloads — efficiency waste, energy consumption, and resource behaviour were all invisible.
    what
    Developed a recommendation system that detects idle workloads, overprovisioned containers, and temporal waste patterns — each with dollar amounts attached. Quick Wins are ranked by estimated monthly savings and ship with copy-paste kubectl commands.
    how
    Built profiling pipelines that compare workload efficiency across deployment versions (cost, energy, CPU, memory, SCI score), surface regional cost optimization opportunities, and generate ready-to-use kubectl commands for immediate savings.
03 / 07 Jan 2024 - Oct 2026
rtCamp

rtCamp ↗

DevOps Engineer

  1. .01

    Enterprise IaC & Migration — Cox Automotive

    why
    S3 buckets, CloudFront distributions, and WAF policies for prod, dev, and staging environments were created manually via aws console and have some legacy settings
    what
    Migrated and standardized S3, CloudFront, and WAF configurations across all three environments into Terraform. with 0 drift terraform drift and 0 downtime.
    how
    Established a multi-environment GitOps workflow with a rigorous Plan-Review-Apply SOP, eliminating click-ops and achieving consistent, repeatable infrastructure across prod, dev, and staging.
  2. .02

    Internal Server and Site Log reporting

    why
    More than 10 servers hosting over 200 WordPress, Frappe, and custom sites lacked centralized log collection and actionable reporting, increasing the risk of missing fatal errors and production issues.
    what
    Built a centralized log intelligence pipeline using Fluent Bit, OpenTelemetry Collector, Loki, Grafana, and an AI agent harness. The system normalizes logs from multiple sources, deduplicates routine noise such as rate-limit events, and produces hourly summaries of application problems.
    how
    Standardized incoming logs into a common OpenTelemetry schema before storing them in Loki, then connected the data to an MCP-enabled AI agent for reporting. Reduced unnecessary tool calls and token usage by downloading and deduplicating logs only when needed, while delegating log processing to efficient UNIX command-line workflows. This made analysis nearly 2x faster and reduced token usage by 70%, giving developers a single view of active incidents including credential leaks, OOM kills, and fatal application errors. Connected Loki to Grafana through a Cloudflare Tunnel.
  3. .03

    Internal CI/CD with GitHub ARC Runners

    why
    Server-hosted GitHub runners polluted host environments, introduced dependency and networking conflicts, and required brittle custom scripts for workflows such as wp-env and Docker Compose. The setup also lacked runner-based autoscaling and cost control.
    what
    Migrated the self-hosted runner platform to Kubernetes-based GitHub Actions Runner ScaleSets with monitoring and autoscaling support.
    how
    Gradually shifted private repositories to the new runner platform with zero CI/CD downtime, updating the necessary GitHub Actions workflows to simplify execution. The migration reduced unexplained pipeline failures, improved maintainability, and brought the developer experience closer to GitHub-hosted ubuntu-latest runners.
  4. .04

    Cloud-Native Scaling & Observability — Global FinTech

    why
    A monolithic Frappe/ERPNext platform was buckling under high-concurrency traffic with no visibility into where production bottlenecks occurred.
    what
    Migrated the platform to a distributed Kubernetes cluster and implemented full-stack OpenTelemetry instrumentation.
    how
    Decomposed the monolith into individual services and migrated to Kubernetes — gaining autohealing, rolling deployments, and the broader ecosystem benefits. Layered OTel logs, traces, and metrics across the stack while deliberately keeping the architecture as simple as possible to reduce operational overhead.
  5. .05

    Cloud FinOps & Cost Engineering

    why
    Memory-intensive background jobs and cron workloads were running on always-on instances, inflating cloud spend without any performance benefit.
    what
    Achieved a 20% reduction in cloud OpEx across Kubernetes compute.
    how
    Engineered specialized Node Groups with Spot Instances and implemented scale-to-zero logic for background and cron workloads, eliminating idle resource waste while maintaining throughput.
  6. .06

    Product Engineering — EasyDash / EasyEngine

    why
    Manual WordPress/PHP deployment processes were slow and error-prone, blocking a commercial product launch.
    what
    Co-developed a high-scale Cloud Provisioning Engine for dash.easyengine.io.
    how
    Built the automated backend with Python, Terraform, and Ansible — enabling rapid deployments that generated $200+ in subscription revenue within 60 days of launch.
  7. .07

    Developer Experience & CI Optimization

    why
    Shared CI runners were creating queue bottlenecks and long wait times that disrupted engineering flow across teams.
    what
    Optimized GitHub Self-Hosted Runners across the organization.
    how
    Applied resource-aware labeling and multi-container environments, drastically reducing CI/CD wait times and improving overall build reliability.
  8. .08

    MariaDB Operations & Disaster Recovery

    why
    Database maintenance and recovery required coordinated procedures for replication, backups, and safe service transitions.
    what
    Contributed MariaDB operational automation and runbooks covering primary/replica setup on Azure, disk snapshots, an independent prepared mariadb-backup recovery path, monitoring, binlog handling, and audit logging.
    how
    Documented replication-aware maintenance and GTID-aligned recovery procedures, including read-only transitions, query draining, replica-lag checks, ordered shutdown/startup, and post-change health verification.
  9. .09

    Cross-Platform Access-Offboarding Automation

    why
    Departing-user access spanned infrastructure and application platforms, making manual cleanup repetitive and easy to miss.
    what
    Contributed Python automation for cloud-server discovery, SSH-key removal, WordPress role removal, and membership cleanup across GitHub, managed WordPress, and Cloudflare.
    how
    Implemented dry-run support, atomic SSH-file updates, and post-change verification, while preserving WordPress user accounts and their content during role removal.
  10. .10

    GitHub Actions Runner Governance

    why
    Runner-selection policy across private repositories needed repeatable checks rather than manual workflow reviews.
    what
    Contributed scheduled workflow scanning, approved runner-label validation, automated remediation pull requests, and notifications for eligible private repositories.
    how
    Scoped scans to non-archived, non-mirror private repositories and documented runner selection and failure-safe cleanup. Supported multiple approved runner labels rather than enforcing a single runner type.
  11. .11

    Email Delivery & Transport-Encryption Observability

    why
    Email operations needed visibility into delivery problems, failed encryption sessions, and transport-policy reporting gaps.
    what
    Contributed scheduled TLS-RPT processing with Gmail API retrieval, compressed JSON attachment parsing, deterministic aggregation, and conditional alerts for encryption failures and reporting anomalies.
    how
    Added SES bounce and complaint reporting with validated AI summaries and a data-only fallback, keeping authoritative counts separate from model-generated interpretation.
  12. .12

    Scheduled Object-Storage-to-Loki Log Ingestion

    why
    Managed WordPress edge and origin logs stored in S3 needed to be accessible through the existing Grafana/Loki investigation workflow.
    what
    Contributed scheduled S3-to-Loki ingestion with manual date and log-type selection, site-based processing, and structured source metadata.
    how
    Preserved source timestamps and stable stream labels, with batching, retries, temporary-file cleanup, and non-zero exit status on ingestion failure. Retained a scheduled workflow rather than adding a more complex event-driven service.
04 / 07 Feb 2025 - Apr 2025
Viamagus

Viamagus ↗

DevSecMLOps Consultant

  1. .01

    API Security Testing with OWASP ZAP

    why
    Client web applications and APIs had undetected vulnerabilities posing real regulatory and business risk.
    what
    Implemented automated API security testing using OWASP ZAP.
    how
    Deployed a custom proxy to surface SQL injection, MITM, and other OWASP Top 10 risks across the application surface, giving the team actionable findings.
  2. .02

    Continuous Vulnerability Scanning with Snyk

    why
    Security issues were caught late in the development cycle, making remediation expensive and disruptive.
    what
    Integrated Snyk into CI/CD for shift-left vulnerability scanning.
    how
    Enabled detection and remediation of dependency and code vulnerabilities at pull request level, before they ever reached production.
  3. .03

    LLM Containerization & Performance Benchmarking

    why
    The team had no baseline data on containerization efficiency, making optimization decisions purely speculative.
    what
    Optimized internal LLM-based projects for containerization and produced detailed performance benchmarks.
    how
    Measured image size, network throughput, and disk I/O to give the team a clear, data-backed picture of real-world efficiency tradeoffs.
  4. .04

    Scalable LLM Deployment on AWS

    why
    LLM inference at scale required elastic capacity and strict network isolation without exposing endpoints publicly.
    what
    Deployed vLLM and Ollama on AWS for production-scale LLM inference.
    how
    Used Auto Scaling Groups and VPC PrivateLink to deliver secure, elastic inference capacity without public endpoint exposure.
05 / 07 Jul 2022 - Jul 2025
kli - ksctl

kli - ksctl ↗

Creator

  1. .01

    High-Performance CLI Engineering

    why
    Provisioning Kubernetes clusters across multiple cloud providers required deep provider expertise and many tedious manual steps.
    what
    Architected kli — a multi-cloud CLI using Cobra and Viper — to abstract the full Kubernetes lifecycle.
    how
    Built a pluggable architecture with robust configuration management, enabling single-command cluster creation and teardown across AWS, Azure, and Civo.
  2. .02

    Custom Kubernetes Controllers

    why
    Cluster state could drift between local CLI metadata and remote cloud infrastructure, causing hard-to-debug inconsistencies.
    what
    Developed custom Kubernetes controllers and reconciliation logic using client-go.
    how
    Ensured idempotent state management that reliably kept local metadata and remote infrastructure in sync.
  3. .03

    Automated Addon & Helm Integration

    why
    Manual post-provisioning addon setup was repetitive, error-prone, and delayed clusters reaching a ready state.
    what
    Automated deployment of core cluster components — CNI, Storage Classes, and Ingress — as part of provisioning.
    how
    Leveraged native Helm SDKs and Go client packages to install components consistently on every cluster without manual intervention.
  4. .04

    Secure Distribution & Artifact Signing

    why
    Users needed confidence that CLI binaries were tamper-free and reliably distributed across all platforms.
    what
    Implemented secure artifact signing and automated multi-platform binary releases.
    how
    Integrated Cosign for signing and GitHub Actions for cross-platform builds, delivering a trusted and seamless developer distribution experience.
06 / 07 Mar 2022 - Jul 2024
Kubesimplify

Kubesimplify

Ambassador

  1. .01

    Technical Content & Education

    why
    Developers onboarding to cloud-native lacked approachable, practical resources for Kubernetes and Go.
    what
    Authored blogs and tutorials on Kubernetes and Go.
    how
    Published accessible technical content aligned with real-world use cases for the Kubesimplify community.
  2. .02

    Live Cloud-Native Sessions

    why
    Complex topics like cloud-native architecture benefit more from interactive live walkthroughs than static articles.
    what
    Conducted Twitch live sessions on cloud-native development and Golang best practices.
    how
    Engaged the community through live demos and real-time Q&A, making advanced topics approachable.
  3. .03

    Open Source Maintenance

    why
    The CNCF ecosystem needed reliable, community-maintained tooling for Kubernetes workflows.
    what
    Maintained and improved Ksctl, contributing to CNCF-aligned open-source tooling.
    how
    Contributed code, fixes, and documentation to keep the project active and aligned with community standards.
07 / 07 Mar 2023 - Sep 2023
Viamagus

Viamagus ↗

DevOps Intern

  1. .01

    Kubernetes Migration to EKS

    why
    The existing cluster lacked the scalability and traffic management needed for production workloads.
    what
    Supported Kubernetes migration to EKS with NGINX Gateway API for traffic management.
    how
    Configured EKS and set up Gateway API routing rules to handle production-grade traffic reliably.
  2. .02

    Application Containerization

    why
    Node.js applications weren't containerized, limiting deployment consistency and portability across environments.
    what
    Dockerized Node.js applications and migrated the reverse proxy from Apache to NGINX.
    how
    Wrote Dockerfiles for each service and reconfigured NGINX for improved routing and performance.
  3. .03

    CI/CD Automation

    why
    Manual Jenkins job setup for new projects was repetitive and slowed pipeline delivery for the team.
    what
    Automated Jenkins job creation using a CLI tool.
    how
    Built a CLI automation layer that provisioned jobs from config, eliminating manual setup overhead.
  4. .04

    Security & Observability Integration

    why
    The team had no visibility into vulnerabilities or runtime metrics before production deployments.
    what
    Integrated Snyk for vulnerability scanning and Prometheus for real-time observability.
    how
    Configured Snyk in CI for early detection and set up Prometheus scraping for live service metrics.
  5. .05

    Infrastructure & SSL Automation

    why
    Manual SSL provisioning and repository management were error-prone and time-consuming at scale.
    what
    Migrated repositories to AWS CodeCommit and automated SSL provisioning.
    how
    Set up Let's Encrypt automation for SSL certificates alongside the CodeCommit migration.
/how-i-work

Good systems reduce cognitive load.

01

Lead with the outcome

Start with the user problem and the impact, then explain the system design, architecture, and trade-offs. Technical depth is most useful when it helps people make better decisions.

02

Make the reliable path the easy path

Good system design removes repeated judgment calls: boring deployments, observable services, simple CI/CD, and automation that behaves predictably.

03

Use AI with a human in the loop

AI is a strong research and thinking partner when it can see the right telemetry and context. High-stakes changes still require human review, explicit boundaries, and ownership.

Workspace
fig.03 / workspace shutter open
/achievements

A few things I'm proud of.

  • 01 OSS contributions to Kubernetes, CNCF TAG Green, Kubescape, Monokle, and more 2024
  • 02 Invited to GitHub Maintainers repo 2024
  • 03 PR Wrangler in sig-docs (Kubernetes) 2024
  • 04 Member of Kubernetes and Kubernetes-Sigs Organization 2023
  • 05 Winner of Napptive + WeMakeDevs Cloud Native Hackathon (Track 2) 2022
/writing

Learning in public

I write from experiments rather than certainty: starting with a question, trying something unfamiliar, then going deep on what actually happened. The collection spans system design, AI, Kubernetes, databases, Kubmin, home-server experiments, and quantum computing.

Browse the collection ↗
/cta · the next move

Pick a thread.
Let's pull on it together.

DevOps consulting, Kubernetes audits, Go service design, or just a 30-minute sanity-check on your platform — start with whatever feels right.

Send a brief → LinkedIn ↗ Resume PDF ↗