huzaifa@gill: ~/portfolio — zsh
whoami

Huzaifa Ahmad Gill

cat role.txt
Backend & Cloud / DevOps engineer
cat pitch.txt

I build backend services and the infrastructure they run on — gRPC/REST APIs in Go, Python and Node, deployed on Kubernetes and wired together with Terraform and ArgoCD.

cat also.txt

A lot of my work is AI/ML too — agentic LLM/RAG systems and end-to-end MLOps (training + serving). Backend and infra are my core; they're what make the ML actually ship.

open to backend / infra / DevOps roles
→ see the work download cv ↓ github ↗ linkedin ↗ email
## about

I'm most at home on the backend and infrastructure layer — designing services, connecting them over gRPC and message queues, and getting them to run reliably on Kubernetes. I like the unglamorous parts: schemas as contracts, CI/CD that can't foot-gun you, cost controls, and observability you can actually debug with.

Backend + cloud/DevOps is my core. A big part of my work is also AI/ML — agentic LLM/RAG systems and end-to-end MLOps for training and serving models — and the Kinetics platform below is where the infra and the ML fully meet. If you want someone who'll happily live in Terraform, Helm charts and gRPC contracts and understands the ML systems running on top, that's the fit I'm aiming at. Frontend I'll pick up when a project needs it.

// education
BSc (Hons) Computer Science — University of Hertfordshire
Oct 2022 – Aug 2025 · Upper Second Class Honours (2:1)

## stack — what I actually reach for
languagesGo · Python · TypeScript/Node · Java
backendgRPC · REST · GraphQL · FastAPI · NestJS · Spring Boot
cloudAWS (EKS · EC2 · RDS · Lambda · S3 · MSK) · Terraform · Terragrunt
version controlGit · GitHub · GitLab
orchestrationKubernetes · Helm · ArgoCD · Argo Rollouts · KEDA
ci / deliveryGitHub Actions (keyless OIDC) · Docker · Atlantis
data / msgPostgreSQL · Redis · Kafka · MongoDB
observabilityPrometheus · Grafana · OpenTelemetry · Loki
ai / ml / mlopsPyTorch · MLflow · SageMaker · LangGraph/LangChain · RAG · agents
alsoReact · Vue · Neo4j · ASP.NET Core
## experience

Machine Learning Intern @ Red Buffer

Apr 2026 – May 2026 · onsite
  • Built FlowForge, an AI planning platform that turns client proposals into validated timelines, Gantt charts and architecture diagrams via a LangGraph multi-agent pipeline.
  • Architected a 4-agent LangGraph workflow with self-healing retry loops that automatically recover from invalid or malformed agent outputs.
  • Integrated HuggingFace Inference API and Together AI with LangChain prompt optimisation; shipped a Dockerised FastAPI service backed by PostgreSQL (SQLAlchemy + Alembic) for output versioning.
→ FlowForge repo ↗
## work — two I'm proud of, then the rest

01Multi-Tenant SaaS Platform

★ flagship · in dev

A multi-repo microservices system: every service lives in its own repo with its own CI/CD and deploy lifecycle. Services talk over gRPC using typed contracts from a central proto repo (Buf registry) as the single source of truth. The whole thing runs on an AWS environment provisioned in Terraform and shipped by ArgoCD GitOps across dev/staging/prod.

saas-platform/
├─ istio-gateway        mTLS ingress · PeerAuthentication
├─ api-gateway (Go)     go-kit · JWT · Redis rate-limit · circuit breaker
│    ├─GraphQL─► auth-service          Node · Express 5 · Prisma · PostgreSQL
│    ├─GraphQL─► subscription-service  NestJS · TypeORM · Kafka publisher
│    └─REST────► billing-service       Java · Spring Boot · Stripe
│                     └─gRPC─► subscription-service
├─ usage-service (Python)  Airflow DAGs · ChromaDB RAG   ◄─Kafka
├─ proto/       Buf schema registry — gRPC contracts
├─ infra/       Terraform · Terragrunt · EKS · RDS · MSK · Atlantis (PR-driven)
├─ security/    OPA/Rego · Conftest · Trivy — policy gates + image scans in CI
└─ delivery/    ArgoCD · Argo Rollouts (canary) · KEDA · Crossplane · CUE
GoNestJSSpring BootNodePython gRPC · BufGraphQLKafka / MSKIstio mTLS Keycloak OIDCArgoCDArgo RolloutsKEDA CrossplaneAtlantisEKS · TerraformTerragruntOpenTelemetry OPA · RegoConftestTrivy
→ github ↗

02Kinetics MLOps Platform — HyperPod on EKS

★ flagship

End-to-end MLOps platform for training a video action-recognition model (PyTorch CNN-LSTM) on Kinetics-400, running on SageMaker HyperPod orchestrated by EKS — with cost controls built in from day one. GPUs default to scale-to-zero: Karpenter only provisions a Spot GPU node when a job is pending, then consolidates back to zero. Training checkpoints to S3 so it auto-resumes after a Spot interruption.

$0idle-GPU cost
~−60%via Spot
12Terraform modules
2repos (infra + CD)
kinetics-mlops/
├─ training     PyTorch CNN-LSTM · DDP · AMP bf16 · torch.compile
│    └─ HyperPodPyTorchJob   torchrun on EKS 1.34 + SageMaker HyperPod
├─ gpu          Karpenter scale-to-zero · Spot-first · auto-resume from S3
├─ storage      S3 ◄─lazy-load─ FSx for Lustre · DVC manifests
├─ tracking     SageMaker MLflow · Model Registry (approval-gated)
├─ serving      SageMaker Endpoint + self-hosted FastAPI
├─ infra/       Terraform · Terragrunt · EKS Pod Identity · Client VPN (SAML)
├─ ci/          GitHub Actions · keyless OIDC · ECR (train + infer)
├─ security/    OPA/Rego · Conftest · Trivy · Checkov — policy gates + scans
├─ observ/      Prometheus · Thanos · Grafana · DCGM · OTel
└─ delivery/    ArgoCD app-of-apps · GPU jobs on manual sync
PyTorch CNN-LSTMDDP · AMP bf16Kinetics-400 SageMaker HyperPodEKS 1.34Karpenter (Spot) FSx LustreMLflowFastAPI servingDVC TerraformTerragruntArgoCDOIDCDCGM · Prometheus ThanosOPA · RegoConftestTrivyCheckov
→ kinetics-pipeline ↗ → gitops repo ↗

03Voice AI Web Application

⟳ ongoing

Real-time voice assistant that handles the full loop — Whisper + Silero VAD transcription, LLaMA 3.3 reasoning, XTTS v2 synthesis — decoupled through a RabbitMQ async queue for low latency. Async Django over gRPC + WebSockets, PostgreSQL via PgBouncer, Redis caching, deployed on Kubernetes with Helm. Currently building out a full observability + DevSecOps stack around it.

Django (async)Next.jsWhisper · Silero VAD LLaMA 3.3XTTS v2gRPC · WebSocketsRabbitMQ PgBouncerRedisKubernetes · HelmNVIDIA MicroK8s PuppetPrometheus · ThanosGrafanaLoki OpenTelemetry
→ github ↗

04Multi-Agent RAG System

LangGraph-orchestrated multi-agent RAG. A planner routes queries across three agents — RAG (LLaMA 3.3 70B via Groq + ChromaDB), Web Search (Tavily + Wikipedia), and Memory — with a ReAct-style replan step, then a Claude aggregator synthesises the answer. Each agent runs as its own MCP microservice over HTTP. Vue 3 + FastAPI.

LangGraphMCP microservicesChromaDBFastAPIVue 3LangSmith
→ github ↗

05Online Banking System

Spring Boot microservices with an event-driven core over Kafka, a multi-database strategy (PostgreSQL + MySQL + MongoDB), and full observability with Grafana, Prometheus, Loki and Tempo. A good exercise in wiring services together properly.

Spring BootKafkaPostgreSQL · MySQL · MongoDBGrafana · Loki · Tempo
→ github ↗

06Crypto Currency Web Application

Crypto trading platform in Go with gRPC microservices, deployed AWS-native on EKS with RDS + ElastiCache, GitOps via ArgoCD, and full IaC through Terraform + Helm.

GogRPCEKSRDS · ElastiCacheArgoCDTerraform · Helm
→ github ↗

07Logistics Supply-Chain Analyzer

Supply-chain engine on ASP.NET Core (.NET 10) + Neo4j, built with Clean Architecture & CQRS. Models warehouses and routes as a graph and exposes a JWT-secured, rate-limited REST API for shortest-path analytics and delivery-risk scoring. Redis caching, Kafka + RabbitMQ messaging, Testcontainers integration tests.

.NET 10Neo4jCQRS · MediatRRedisKafka · RabbitMQTestcontainers
→ github ↗

08FlowForge

Feed it a client proposal and a LangGraph multi-agent pipeline returns timelines, Gantt charts and technical diagrams — with self-healing retry loops that fix invalid outputs before delivery. FastAPI backend, PostgreSQL session store.

LangGraphHuggingFace APIFastAPIPostgreSQLSelf-healing agents
→ github ↗
more projects — the rest of the lab (9)
Cloud-Based RAG Architecture ↗Serverless RAG on AWS — Lambda ingest, DynamoDB vectors, Bedrock, Terraform.
MCP Code Advisor Web App ↗Full-stack AI code advisor over MCP. EC2 + Route 53 + TLS, Terraform + Ansible, Apache2, Linux Ubuntu, GH Actions.
Agentic RAG (Multi-LLM) ↗LangChain agents that pick which model (LLaMA 3.1/3.2, GPT-4.1-mini) to call per sub-task.
Multimodal RAG ↗RAG over text, images and structured data with cross-modal retrieval.
ETL Pipeline ↗Airflow ETL + MLflow tracking + DVC versioning + Prometheus observability on S3.
Video Streaming & Event Pipeline ↗Serverless video ingest — Kinesis + Lambda + Redshift + OpenSearch, X-Ray tracing.
Stock Price Forecasting (FYP) ↗LSTM forecasting model behind a Flask app on live market data. Final-year project.
Sequence-Based Neural Network ↗Spatio-temporal video prediction — CNN+LSTM vs ConvLSTM2D on FlyingThings3D.
A3C Reinforcement Learning Agent ↗Async Advantage Actor-Critic on Gymnasium, distributed across actor-learner threads (PyTorch).
## contact

Open to backend, cloud or DevOps roles — full-time, contract, or a genuinely interesting infra problem. Easiest to reach by email.

emailhuzaifaahmad2210@gmail.com
githubgithub.com/huzaifa678
linkedin/in/huzaifa-ahmad-gill
resumecv.pdf ↓