AI engineer · full-stack · ships end-to-endIndia · remote (SG / US / EU hours)

I turn messy AI ideas
into systems that ship.

Most recently at an early-stage telecom AI startup: agents on Kubernetes, production RAG, integrations (Jira, Outlook, WhatsApp), and customer-facing demos — the kind of work that happens when there is not a separate "infra team."

If you are a founder or hiring manager trying to answer "can this person own ambiguous AI work?" — start with Work, then OSS.

View shipped work
Scroll
How I think

Wired to think
like a builder.

"The architecture decision I'm most proud of isn't the one I made — it's the one I talked us out of. We saved three weeks by not building the wrong abstraction."
— SWARM: the refactor we skipped (and why)
01 —
Full-stack ownership
At an early-stage startup there's no "infra person." You either own the whole thing or nothing gets built.
02 —
Speed with judgment
The system that ships in week 2 still needs to be understood in week 10 — by me, or by whoever comes next. Speed without that isn't speed, it's debt.
03 —
Integration-first thinking
LLMs in notebooks aren't products. Products are wired to Jira, Outlook, WhatsApp, and real systems. I've done that wiring — not theoretically.
04 —
Product sense
I know what founders actually need to hear — and how that differs from what sounds technically impressive.
Work · Internship & Projects

Things I've actually built

Scroll · drag
01
Multi-AgentEKS · K8sInternship

SWARM

Most "multi-agent" demos are two LLM calls in a trenchcoat. SWARM runs each agent in its own Kubernetes pod, spawned on demand, wired to real enterprise systems — Jira, Outlook, internal ticket queues. Ticket triage, email routing, network fault resolution.

Google ADKLangGraphTerraformLiteLLMEKS
Telecom AI startup · Singapore
Email about this project
02
AI Product100k+ scaleInternship

AutoCLM

Telcos blast the same SMS to 100,000 people and call it personalisation. AutoCLM is the alternative — continuous 1:1 WhatsApp conversations, each customer on their own AI-planned journey arc. Built the full stack: backend, WhatsApp integration, team-facing dashboard, and five realistic customer personas for demo.

LangGraphWhatsApp APIReactFastAPI
Telecom AI startup · CLM / WhatsApp initiative
Email about this project
03
AI InfraHybrid SearchInternship

Info Repo v2

"Close enough" is not acceptable when an AI is quoting plan prices to real customers. I scrapped v1 and rebuilt — hybrid dense+BM25 retrieval, RRF fusion, Cohere reranking, LLM-generated chunk labels. Four content zones, approval loop for all ingested data, soft deletes only.

QdrantBM25+RRFCohereFastAPIMinIO
Telecom AI startup · knowledge / pricing accuracy
Email about this project
Personal projects
04
RAG · InsurancePersonal

Insurance RAG

A retrieval system for insurance policy documents — the kind of dense, jargon-heavy text where standard chunking falls apart. Built to answer real coverage questions accurately, not confidently-wrongly.

LangChainQdrantFastAPIPython
Personal · accuracy-first RAG
Email about this project
05
Go · MicroservicesPersonal

Go gRPC Microservices

Built a distributed system in Go to learn gRPC and GraphQL properly — not a tutorial clone, but something with real service boundaries, protobuf contracts, and an API gateway. The kind of project you build when you want to understand the seams.

GogRPCGraphQLProtobuf
Personal · distributed systems study
Email about this project
More experiments

Not everything fits a card. If something here overlaps what you are building, ask — I will point you to the closest parallel.

WebRTC object detectionMicro ADK frameworkPDF extractorFluxShare · RakshaChakra
Email about other builds
Stack

Everything I've
shipped with

LangGraphGoogle ADKKubernetesTerraformGoPythonTypeScriptFastAPILangGraphGoogle ADKKubernetesTerraformGoPythonTypeScriptFastAPILangGraphGoogle ADKKubernetesTerraformGoPythonTypeScriptFastAPI
QdrantBM25+RRFCohere RerankOpenRouterLiteLLMAWS EKSMinIOPostgresQdrantBM25+RRFCohere RerankOpenRouterLiteLLMAWS EKSMinIOPostgresQdrantBM25+RRFCohere RerankOpenRouterLiteLLMAWS EKSMinIOPostgres
ReactgRPCGraphQLWebRTCJira WebhooksSvixAgentMailWhatsApp APIReactgRPCGraphQLWebRTCJira WebhooksSvixAgentMailWhatsApp APIReactgRPCGraphQLWebRTCJira WebhooksSvixAgentMailWhatsApp API
GitHub Contributions
Fetching GitHub data
Open source

I contribute to
things that matter.

Merged PRs on real codebases — features, fixes, and reliability — not drive-by typo PRs for green squares. Every repo below reviewed my code against engineers worldwide.

docker/docker-agent
Primary focusActive contributor

A production AI agent codebase used by developers worldwide — CI on Linux and macOS, real code reviews, engineers from across the ecosystem. 15+ merged PRs, not one-liners.

  • ↳Background job management for shell toolset (run, list, view)
  • ↳Filesystem tools that respect .gitignore rules across all operations
  • ↳Auto directory creation in write_file for nested paths
2.8k
Stars live from GitHub
Contributor rank: Top 10Merged PRs: 15+Review cycles: real engineers
View merged PRs
akuity/kargo
~6k
GitOps · CD
stacklok/toolhive
1.7k
MCP · Security
vllm-project/aibrix
~2k
LLM Infra
kubernetes/autoscaler
~8k
Kubernetes
24+
Merged PRs across OSS repos
Top 10
Contributors docker/docker-agent
6+
Production repos real review cycles
Contact
Let'sbuildsomethingreal.

Send the problem in plain language — what is broken, what you have tried, what "good" looks like. I will reply with how I would tackle it (and whether I am the right person). If it is a fit, we will find time to talk.

deepam02goyal@gmail.com→LinkedIn→GitHub→
Availability snapshot
StatusOpen to offers
TypeFull-time · Contract
LocationIndia / Remote
NoticeImmediate
Ideal team size2–20 people
What I won't doCRUD apps. Pure PM roles.
Email deepam02goyal@gmail.com