■■ Engineering Tech Stack

Reference
Languages · Frameworks · DevOps · AI/ML — ZenithGPT Knowledge Base v2.0

1. Programming Languages — When to Use What

Language Strengths Primary Use Cases

Python Rapid dev, huge ecosystem, readable AI/ML, data science, scripting, backend APIs

Go (Golang) Fast compilation, low latency, goroutinesMicroservices, CLI tools, networking, DevOps

Rust Memory safety without GC, bare-metal speed

Systems, WASM, game engines, embedded, crypto

Java JVM ecosystem, enterprise battle-testedEnterprise backend, Android, big data (Spark)

TypeScript JS + type safety, full-stack Frontend (React/Angular), [Link] backend, monorepos

C++ Maximum performance, hardware control

Games, embedded, HFT, OS kernels, compilers

Kotlin Modern JVM, concise, coroutines Android, server-side (Ktor, Spring Boot)

Swift Safe, fast, Apple ecosystem iOS/macOS apps, Swift on Server

Scala Functional + OOP on JVM Distributed systems (Akka, Spark), big data

Elixir Actor model, fault-tolerant, hot reload Real-time systems, chat, telco, IoT

2. Web Frameworks Comparison

Framework Lang Paradigm Best For

FastAPI Python Async ASGI ML APIs, high-perf REST, auto OpenAPI docs

Django Python Batteries-included Full-stack web, CMS, ORM, admin panel

[Link] [Link] Minimal REST APIs, BFF, real-time with [Link]

NestJS TypeScript Angular-style DI Enterprise Node APIs, GraphQL, microservices

Gin Go Ultra-fast router High-throughput microservices, REST APIs

Spring Boot Java/Kotlin Comprehensive Enterprise Java, Spring ecosystem, cloud-native

Actix-web Rust Actor-based Maximum performance web services

Rails Ruby Convention over config Rapid prototyping, startups, CRUD apps

Laravel PHP Elegant ORM/DI Traditional web apps, e-commerce

[Link] TypeScript SSR/SSG/ISR SEO-critical React apps, full-stack web

3. DevOps & Infrastructure Stack
3.1 Container & Orchestration
Tool Purpose Key Commands / Notes

Docker Container runtime & image build

docker build, run, push, compose up

Kubernetes Container orchestration kubectl apply, rollout, scale, logs

Helm K8s package manager helm install, upgrade, rollback, [Link]

ArgoCD GitOps CD for K8s Declarative, Git = source of truth, auto-sync

Kustomize K8s config management Overlays for env-specific configs, no templates

Istio Service mesh mTLS, traffic management, observability, Envoy sidecar

Linkerd Lightweight service mesh Simpler than Istio, Rust-based, lower overhead

3.2 Infrastructure as Code (IaC)

Tool Language Best For

Terraform HCL Multi-cloud IaC, state management, 1500+ providers

Pulumi Python/Go/TS/Java IaC in real programming languages, same as app code

AWS CDK Python/TS/Java AWS-specific, synthesizes to CloudFormation

Ansible YAML (declarative) Config management, app deployment, agentless SSH

Chef/Puppet Ruby DSL Config management (legacy enterprise environments)

3.3 Observability Stack (The 3 Pillars)

Pillar Open Source Managed/SaaS

Metrics Prometheus + Grafana Datadog, New Relic, CloudWatch

Logs ELK Stack (Elasticsearch, Logstash, Kibana) Splunk,

/ Loki Datadog Logs, CloudWatch Logs

Traces Jaeger, Zipkin, OpenTelemetry Datadog APM, AWS X-Ray, Honeycomb

OpenTelemetry (OTel) is the emerging standard. Instrument once, send to any backend.

3.4 CI/CD Pipelines

Tool Hosted/Self Strength

GitHub Actions Hosted (GitHub) Native GitHub integration, 15k+ marketplace actions

GitLab CI Both Integrated with GitLab, powerful pipeline DSL, self-hostable

Jenkins Self-hosted Maximum flexibility, plugin ecosystem (1800+), enterprise

CircleCI Hosted Fast, Docker-native, orbs reusability

Tekton Self-hosted (K8s) Cloud-native K8s pipelines, GitOps friendly

 4. AI / ML Engineering Stack
Layer Tools / Frameworks Notes

Data Processing Pandas, Polars, Spark, DuckDB Polars = Rust-based, 10–100x faster than Pandas for large data

Feature Store Feast, Tecton, Hopsworks Prevent training-serving skew, reuse features

Model Training PyTorch, TensorFlow, JAX, Scikit-learn PyTorch dominates research; TF for production TFX

Experiment Tracking MLflow, W&B, Neptune, Comet Track params, metrics, artifacts, model versions

Model Serving TorchServe, Triton, BentoML, Ray ServeTriton: NVIDIA multi-framework; Ray Serve: scalable Python

LLM Fine-tuning LoRA, QLoRA, PEFT (HuggingFace) QLoRA: 4-bit quant + LoRA = fine-tune on 1 GPU

RAG Pipeline LangChain, LlamaIndex, Haystack Retrieval-Augmented Generation for grounded LLM apps

Vector Databases Pinecone, Weaviate, Qdrant, pgvector pgvector = PostgreSQL extension, simplest for existing PG setups

LLM Observability LangSmith, Helicone, Arize Trace, evaluate, monitor LLM apps in production

Model Optimization ONNX, TensorRT, vLLM, [Link] vLLM: PagedAttention for high-throughput LLM serving

5. API Design Standards

Aspect REST GraphQL gRPC

Protocol HTTP/1.1 or 2 HTTP/1.1 or 2 HTTP/2

Format JSON (mostly) JSON Protocol Buffers (binary)

Schema OpenAPI/Swagger SDL (strongly typed) Proto files (strongly typed)

Streaming Limited (SSE/WS) Subscriptions Native bi-directional

Perf Moderate Moderate High (binary, multiplexed)

Browser native Yes Yes No (needs grpc-web)

Best for Public APIs, CRUD Flexible client queries, BFF Internal microservices, mobile

6. Security Essentials Checklist

■ Authentication: Use OAuth 2.0 + OIDC. JWT (RS256 > HS256). Short-lived access tokens (15 min),
long refresh tokens.
■■ OWASP Top 10 must-know: Injection, Broken Auth, XSS, IDOR, Security Misconfiguration, SSRF,
Crypto failures.
■ Encryption: TLS 1.3 in transit. AES-256-GCM at rest. Never roll your own crypto.
■■ Zero Trust: Never trust, always verify. mTLS between services. Short-lived service credentials.
■ Least Privilege: IAM roles scoped to minimum needed. Rotate secrets. Use Vault or AWS Secrets
Manager.
■ Input Validation: Validate and sanitize all inputs. Parameterized queries only. Never eval() user
input.
■ Audit Logging: Log all auth events, privilege escalations, data access. Immutable log store.

ZenithGPT Knowledge Base — Tech Stack Reference v2.0 | For AI use only