Solution architect · Platform engineering

I design and build multi-tenant AI platforms.

Multi-tenant AI platforms — architecture through to production, and everything that has to keep running afterwards.

From signup to serving · one tenant’s journey {{ phase }}

Build

GitHub

Actions

Registry

Helm

GKE cluster

ingress · TLS · routing

acme

app
workers
data

globex

app
workers
data

initech

app
workers
data

MongoDB · Qdrant · Redis

+49 more namespaces · nothing shared

Platform

LLM APIs

Chargebee

Monitoring

Network policy

{{ act }}{{ traceText }}

{{ traceIndex }}

{{ latency }} ms p95 {{ rps }} rps 99.9% uptime
12
Engineers led across
backend and frontend
50+
Enterprise tenants on
platforms I architected
99.9%
Uptime across the
platforms I run
9
Years across backend,
platform and security

01Selected work

{{ counter }} · client names withheld

Select a project · 5

01Reading

Multi-tenant chatbot training SaaS

01

Multi-tenant chatbot training SaaS

02Reading

Agentic analytics and dashboarding platform

02

Agentic analytics and dashboarding platform

03Reading

AI interviewer — automated technical screening

03

AI interviewer — automated technical screening

04Reading

Advisor–advisee consultation platform

04

Advisor–advisee consultation platform

05Reading

Industrial workflow automation

05

Industrial workflow automation

{{ title }}

{{ role }}

The problem

What I built

{{ problem }}

{{ built }}

Architecture

{{ layer.t }} {{ layer.line }}

{{ stackLine }}

View the architecture diagram

Outcome

{{ outcome }}

Also delivered

Digital twins of industrial plants · multi-plant single sign-on architecture using CyberArk · IPSec tunnel connectivity into client environments · an AIOps tool extending HolmesGPT to monitor and remediate services across GCP VMs and Kubernetes · MLOps and CI/CD pipelines · a spec-driven development workflow adopted across the engineering team.

A platform in operation

Fifty-plus organisations, one platform, no shared state.

Isolation sits at the namespace and data-store boundary rather than in application code — which is why tenant count grows without the blast radius growing with it.

Load Namespaces Billing

{{ rps }} rps

Peak sustained

autoscaled

52

Tenants

99.9%

Uptime

52 namespaces

ns/tenant-amongo · redis · qdrant
ns/tenant-bmongo · redis · qdrant
ns/tenant-cmongo · redis · qdrant
+ 49 more, one per organisation

3 plan tiers

Enterpriseseats · SSO · SLA
Businessusage caps · overage
Teammetered, self-serve

Entitlements are enforced in the platform, not the billing provider — a downgrade takes effect the moment it is signed.

02How I work

How an engagement runs

Six stages, always in this order. Testing, review and deployment happen inside each build phase rather than being saved for the end.

  1. 01

    Assessment

    Designs and requirements read in full. Scope, effort and risk fixed before anything is quoted.

    Scope matrix / effort model / risk register / assumption log

  2. 02

    Discovery and architecture

    The system defined end to end: data, service boundaries, tenancy, failure modes, performance targets.

    Architecture diagrams / data model / API contracts / auth & tenancy / NFR targets

  3. 03

    Setup and foundation

    Environments as code, pipelines and observability first — then the foundation every feature leans on.

    IaC environments / CI-CD / secrets / observability / RBAC / billing / notifications

  4. 04

    Phased build

    Capability phases in agreed order. Each one closes with a reviewed, tested, deployed release.

    Inside every build phase

    Build Test Review Release

  5. 05

    Launch

    UAT, independent penetration test, load test to target, cutover with a rollback path, full handover.

    UAT sign-off / VAPT report / load-test results / runbooks / credentials

  6. 06

    Operate and evolve

    SLO monitoring, defined escalation, patch cadence, cost review — and the next phases when you need them.

    SLO dashboards / alerting & escalation / patch cadence / cost reviews

Manoj Kumawat
Manoj Kumawat Solution architect

03About

Systems, then platforms, then AI

Nine years moving in one direction — from systems and security engineering into platform architecture and AI.

2013—17 B.Tech Computer Science · IIT (BHU) Varanasi Fundamentals first — systems, networks, security.
2017—22 Samsung R&D Institute, Noida Software Engineer, then Lead Engineer from 2020 — where the standard for production software was set early.
2022 Senior Software Engineer · Cadence Design Systems Multithreaded systems work on EDA tooling.
2022—24 CTO and Co-founder · Tynza Built and ran the platform end to end on AWS and MongoDB.
2024—now Technical Lead · Algo8 AI Senior Backend Developer, then Technical Lead from Oct 2024 — heading the technical team across GCP, AWS and Kubernetes, taking AI platforms from architecture through to production.

04What I work with

Six domains · all in production

Architecture and platform

01
Multi-tenant SaaSDistributed system designConfig-driven developmentSpec-driven SDLCTechnical assessmentEstimation

AI and data

02
RAG systemsAgentic workflowsLLM integrationSemantic modelling layersMLOpsVector search

Backend

03
Node.jsPythonREST and service APIsSQL, NoSQL, time-seriesDatabase design

Frontend

04
ReactComponent systemsDesign-system implementation

Cloud and infrastructure

05
GCPAWSKubernetes and GKEDockerCI/CDAutoscalingCost optimisation

Security and networking

06
SSL/TLSIPSecCryptographyVAPTSSO and identityAccess control

Contact

Let's talk about what you are building.

If you are building a platform, I can take it from architecture to production — or step in wherever it currently stands. Either way it starts with understanding the system properly before anything gets committed to.

A first call is usually thirty minutes: what you are building, where the hard parts are, and whether I am the right person for it.

Multi-tenant chatbot SaaS · production architecture
− {{ zoomLabel }} + Fit Close ✕
Production architecture of the multi-tenant chatbot SaaS on Google Kubernetes Engine

Scroll to pan · Esc or click outside to close