Skip to content

Open to new roles

Shubham Gupta

Backend AI engineer building multi-agent systems, LLM product features and the event-driven infrastructure that keeps them reliable.

Résumé

Built backends at

  • Memorea
  • Simpplr
  • NE Group
  • Credain
  • Idea Usher
  • IIT Hyderabad

How I build AI that holds up in production

Agents that plan before they act

An orchestrator breaks the task down, hands each step to a specialised sub-agent with the right tools, and checks the result before it answers — fewer hallucinations on multi-step work.

Delivery that survives outages

Event-driven pipelines with idempotent consumers, retries and dead-letter queues — so a broker going down at 3am is a log line, not an incident.

Selected work

Tools I built to understand a problem properly — mostly infrastructure that has to keep working when something downstream fails.

QueryLens

Code on request

Natural-language slow query explainer for PostgreSQL, grounded in real execution plans rather than model guesswork. Finding out why a query is slow means reading EXPLAIN ANALYZE output by hand, and an LLM's answer is only useful if it is grounded in the real plan, real table structure and real index availability. QueryLens diagnoses against the actual execution plan, then generates runnable index migrations and query rewrites with before/after cost comparisons — exposed as a REST API and CLI for CI pipelines.

  • Node.js
  • NestJS
  • OpenAI API
  • PostgreSQL
  • Docker

IdleWatch

Code on request

AWS cost intelligence that identifies genuinely idle resources and prices the waste in monthly spend. Idle cloud resources are invisible until the bill arrives, and a point-in-time snapshot cannot tell genuine idleness from a quiet hour. IdleWatch scores waste by estimated monthly cost impact across RDS, Lambda, EBS and S3 using sustained CloudWatch utilisation metrics, then auto-generates Terraform snippets for right-sizing or terminating flagged resources.

  • Node.js
  • AWS SDK
  • PostgreSQL
  • React.js
  • Terraform
  • Docker

More experiments on GitHub

Impact, in numbers

3+
Years shipping

Backend systems across AI assistants, enterprise SaaS, gaming and Web3 banking.

100K+
Concurrent users

Supported after rearchitecting the API gateway of a high-traffic gaming platform.

50%
Delivery latency cut

Replaced a cron-based system with EventBridge and SQS, with near-100% delivery reliability.

10K
Events per second

Real-time Kafka and Redis pipelines with sub-100ms game-state updates at peak.

Where I've worked

Six roles, most recent first.

  • April 2026 — Present

    50%delivery latency cut

    • Rearchitected the message delivery infrastructure for Memorea, a memory-focused AI assistant, replacing a cron-based system with an event-driven pipeline using AWS EventBridge and SQS, cutting delivery latency by 50% and achieving near-100% delivery reliability at scale.
    • Designed and built a multi-agent architecture from scratch to replace a single prompt-response model, implementing a custom agent harness with an orchestrator, specialised sub-agents, and tool-calling, enabling reliable multi-step task execution and reducing hallucinations.
    • Built email as a first-class channel for Memorea, including AI-powered auto-drafting and auto-labelling of incoming emails, and integrated WhatsApp Business API as a primary conversational channel for user interactions.
    • Built collaborative boards and lists within Memorea fully operable through natural language, allowing users to create, update, and manage tasks conversationally via the agent system.
    • Contributed broadly to product direction and backend architecture, working closely with the founding team on system design decisions as the product scaled.

The toolkit

Every tool here has shipped in production — each area traces back to a role above.

OpenAI APIMulti-Agent OrchestrationAgent Harness DesignTool CallingAgentic WorkflowsRAGPrompt EngineeringPythonTypeScriptJavaScriptSQLSolidityNode.jsNestJSExpress.jsREST APIsGraphQLMicroservicesAPI GatewayWebSocketsReact.jsNext.jsApache KafkaRabbitMQ
BullMQAWS EventBridgeAWS SQSPostgreSQLMongoDBRedisJWTOAuth2RBACSSOMulti-Tenant Access ControlAWSEC2S3LambdaDockerGitHub ActionsJenkinsLinuxGitWhatsApp Business APIEmail APIsGmail / SMTP

AI & LLM Engineering. Custom agent harness — orchestrator, sub-agents and tool calling — reducing hallucinations on multi-step tasks.

Languages. Audited production Solidity to ERC-20 / ERC-721 / ERC-1155 standards; Python and TypeScript across services.

Backend & APIs. Microservices for enterprise SSO and multi-tenant access; API Gateway rearchitected for 100k+ concurrent users.

Messaging & Events. 10k+ events/sec at peak with sub-100ms updates; event-driven pipeline cut delivery latency 50%.

Databases & Caching. Query latency down 40%; write throughput +25% and read latency −30% via denormalization and compound indexing.

Auth & Security. Fine-grained RBAC across distributed services; multi-tenant access control for a SaaS platform.

Cloud & DevOps. CI/CD with automated smart contract testing via GitHub Actions — deployment time down 50%.

Integrations. Email as a first-class channel with AI auto-drafting and auto-labelling; WhatsApp Business API for conversational flows.

AI that helps in production, not just in demos.Let’s build yours together.

Open to backend AI, high-concurrency platform and distributed-systems roles.

Email me