Ramkumar Gupta
Applied AI Platform Architect

Build AI products that are observable, secure and cost-efficient.

Lead full-stack to platform-minded architect: RAG/agents, guardrails, OpenTelemetry, SLOs, Terraform/CDK, and FinOps baked into every feature.

Explore Projects Hire Me Resume Snapshot
p95 latency
< 250 ms
security issues
0 critical
cost / request
₹0.02
deploys / week
20+
availability
99.95%
data sources
> 1M docs
• RAG• Prompt Router• Guardrails• OpenTelemetry• SLOs• Terraform• CDK• SQS/SNS• EventBridge• Step Functions• pgvector• Feature Flags• FinOps• Grafana• k6• Playwright• Backstage• RAG• Prompt Router• Guardrails• OpenTelemetry• SLOs• Terraform• CDK• SQS/SNS• EventBridge• Step Functions• pgvector• Feature Flags• FinOps• Grafana• k6• Playwright• Backstage

Core Skills & Stack

Full-stack + AI + platform: what I ship with every project

AI Application Engineering

RAGEval HarnessPrompt RouterGuardrailsVector SearchStreaming

Platform / DX

BackstageIDPGolden PathsCI/CDFeature Flags

Observability & SLOs

OpenTelemetryGrafanap95/p99Error BudgetsTracing

Security

OWASP LLMPII RedactionTenancyTokenizationAudit Trails

Cloud & Infra

AWSTerraformCDKSQS/SNSEventBridgeStep Functions

Data & Stores

PostgrespgvectorDynamoDBRedisS3

Full-Stack Skill Map

Strengths across the stack with real production use

Front-End

  • React & Next.js (App Router), TypeScript
  • State: Redux & modern hooks patterns
  • UI: Tailwind, custom component systems
  • Accessibility & performance budgets
  • Charts, dashboards, localization (i18n)

Back-End

  • Node.js, Express, serverless (AWS Lambda)
  • Auth (OAuth, Azure AD), RBAC, audit trails
  • Resilient APIs: retries, idempotency, queues
  • ETL & data migrations (Aurora, DMS)
  • Observability: logs, metrics, traces

Data & Storage

  • MongoDB, PostgreSQL, DynamoDB, Redis
  • File/object storage (S3/Cloud Storage)
  • Vector search (pgvector), indexing at scale
  • Schema design, performance tuning

Cloud & DevOps

  • AWS (Lambda, API Gateway, IoT, SQS/SNS)
  • GCP (App Engine, Cloud Functions, Pub/Sub)
  • CI/CD (Jenkins, GitHub), IaC (Terraform/CDK)
  • SLOs, on-call, incident playbooks
Download Resume (PDF)

Project Highlights

A taste of outcomes and trade-offs from recent work

Consumer Apps at Scale

  • Led development of web/mobile apps that improved sales and user satisfaction
  • Localization across frontend & backend; optimized user journeys
  • Live store tracking with AWS IoT; rigorous code reviews

Analytics & IoT Dashboards

  • Architected admin dashboards over large datasets; designed RBAC
  • Cloud functions for image optimization; export of large datasets
  • Built visualizations (Canvas/React) and performance-tuned APIs

Healthcare & Imaging

  • Integrated DICOM viewers; privacy and access control first
  • Built dose tracking & protocol management modules
  • TypeScript + React with robust forms and validation

Data Migration & Serverless

  • Migrated TBs of data to Aurora Serverless with DMS
  • Implemented serverless services with strong auth & localization
  • Search/indexing, caching, and cost-aware architectures

Projects

Interactive cards with motion & tilt

RAG Service (E2E)

Demo Repo

Production-ready retrieval-augmented generation over product docs with evals, caching, and cost/latency panels.

  • pgvector + hybrid search
  • RAG evals pipeline
  • Prompt-injection tests
  • OTel traces end-to-end
AIRAGpgvectorOpenTelemetrySecurity

AI Guardrails Lab

Demo Repo

OWASP LLM Top-10 demo with toggleable mitigations: injection isolation, output validation, PII redaction.

  • Red teaming harness
  • Policy & content filters
  • Audit logs
LLM SecurityOWASPPolicies

Order Orchestrator (Idempotent)

Demo Repo

EventBridge + Step Functions + SQS orchestration for preorder workflow with idempotency, retries, DLQ, and circuit breaker.

  • Exactly-once-like semantics
  • Compensation steps
  • Chaos tests
AWSEventBridgeStepFunctionsResilience

Observability Golden Path

Demo Repo

Drop-in template wiring logs, metrics, and traces across Next.js + Node + DB. One golden dashboard per service.

  • p95/p99 SLOs
  • Trace-based alerts
  • Service map
OpenTelemetryGrafanaSLO

IDP Service Template

Demo Repo

Create-a-service generator with CI/CD, Terraform modules, SLOs, security checks, and cost budgets by default.

  • Backstage catalog
  • Golden paths
  • Policy as code
PlatformTerraformBackstageCI/CD

FinOps Scorecard

Demo Repo

Unit economics for services: request-level cost (tokens + infra), autoscaling, and Savings Plans recommendations.

  • Right-sizing
  • Load tests (k6)
  • Waste reduction
FinOpsk6Autoscaling

Feature Flags & Experiments

Demo Repo

Multi-tenant feature flag service with SDK, audit trail, and staged rollouts/canaries for safe experimentation.

  • Global cache
  • Rollout policies
  • Tenancy
PlatformFlagsExperiments

Search-at-Scale

Demo Repo

Hybrid semantic + BM25 search across ~1M synthetic docs with sharding, query routing, and warm caches.

  • Consistent hashing
  • Shard balancer
  • Perf budgets
SearchShardingScaling

Secure Data Boundary Gateway

Demo Repo

Upload -> scan (DLP/AV) -> store -> signed URLs with PII redaction and strict tenancy boundaries.

  • Least-privilege IAM
  • Egress controls
  • Audit trails
SecurityDLPTenancy

AI E-commerce Assistant

Demo Repo

Next.js storefront + Node APIs + AI copilot (RAG + tools: inventory, orders) with OTel, SLOs, and cost controls.

  • Tool-use router
  • Eval harness
  • Latency/cost targets
AINext.jsSLOCost

Impact Metrics

What I optimize and report for every system I lead

p95 latency
< 250 ms across core APIs
security findings
0 critical / 0 high
cost per request
₹0.02 (AI) | ₹0.005 (non-AI)
deploy cadence
20–30 per week
availability
>= 99.95%
scale
> 1M docs indexed

Open Source & Community

Small, consistent contributions where I build and learn

OpenTelemetry

Plugin samples + docs improvements; trace context propagation utilities for Node/Next.

Backstage / IDP

Templates and example golden paths for Node services with SLOs and budgets.

Case Studies (1-minute reads)

Trade-offs with measurable results

Consistency vs Availability (Payments)

  • Problem: Duplicate payment attempts under retry storms.
  • Action: Chose CP path + idempotency keys + exactly-once-like orchestration.
  • Result: Charge duplicates → 0; p95 auth latency 220ms; error rate 0.2%.

FinOps Hotspot → Unit Economics

  • Problem: LLM endpoint cost spikes during peak traffic.
  • Action: Prompt compression + caching + model routing + autoscaling.
  • Result: Cost/request down 48% with stable p95 and quality.

Let's build something

I am open to Lead/Principal roles and consulting

Email Me GitHub LinkedIn