AI Developer • Software Engineering

Turning Complexity into Clarity.

XGT builds intelligent, reliable, and secure software—combining modern AI with battle-tested engineering.

See Our Work What We Do
• Model engineering (RAG, LLMops) • Backend at scale (Go/Python) • Cloud & Data Infra

Who We Are

We are an engineering-first team specializing in AI systems, backend platforms, and developer tooling. Our approach blends rigorous architecture with delightful UX—so your product is both powerful and a joy to use.

10+
AI & Data Projects
24/7
Reliability Mindset
∞
Curiosity & Craft

Core Stack

  • • Go / Python / Java / Oracle / PostgreSQL (Aurora/RDS) / Redis · gRPC / REST
  • • Vector search: Milvus / Weaviate / FAISS; hybrid retrieval (BM25 + dense) & reranking
  • • LLM & Training: PyTorch, TensorFlow, JAX; LoRA/QLoRA; Hugging Face
  • • Inference: vLLM, Triton Inference Server, ONNX Runtime, TensorRT; gateway & caching
  • • Data & MLOps: Airflow/Prefect, Kafka/Debezium, MLflow/Weights & Biases, Feast (feature store)
  • • Front-End: Three.js · React · Tailwind · Vite/Next
  • • Edge/Mobile AI: TensorFlow Lite, Core ML, ONNX Runtime Mobile, MediaPipe, ExecuTorch
Security by design · Privacy first · Observability baked in

What We Do

AI Systems

R&D → Prod

RAG pipelines (hybrid + reranking), eval harness, agents dengan tools aman, cost-aware gateway (caching/rate-limit), dan hosting model (vLLM/Triton/ONNX) ready to launch.

PyTorch TensorFlow LangChain LlamaIndex LangGraph DSPy Hybrid Search Reranking vLLM Triton ONNX Runtime Ragas Semantic Cache Guardrails

Backend Platforms

Zero-to-Scale

Designing APIs and data layers with strong typing, authz, caching, and principled performance engineering.

Go Python Postgres Oracle Redis gRPC OpenTelemetry

Edge & Mobile AI

On-device

On-device inference (quantized) for privacy & low-latency apps.

TFLite Core ML ONNX Mobile MediaPipe ExecuTorch

How We Work

  1. 1. Discover
    Problem fit, success metrics, risks.
  2. 2. Architect
    Design doc, security & cost plan.
  3. 3. Pilot/PoC
    Prototype with evals & benchmarks.
  4. 4. MVP/MLP
    Ship value fast with guardrails.
  5. 5. Harden
    Observability, stress, pen-test.
  6. 6. Operate
    SLOs, cost controls, hand-off.
Typical: Proposal ≤72h • PoC 2–4 weeks • API P99 ≤800ms

Selected Work

RAVEN

RAVEN

AI

Bringing intelligence, case registers, and AI-driven insights into one unified platform.

OPTRIX

Optrix

Three.js

Next-generation mobile face recognition app built for intelligence and law-enforcement operations.

Trust & Security

Data Handling
  • • TLS everywhere; KMS at-rest (AES-256)
  • • Least privilege, per-service secrets
  • • Private networking/VPC peering
Compliance
  • • GDPR/PDPA-ready, DPA & NDA templates
  • • Optional 3rd-party pen-test
  • • Data residency: SG/ID
AI Safety
  • • Prompt-injection defenses & allow/deny lists
  • • PII redaction & content filters
  • • Model evaluation harness & red-teaming
  • • Versioned rollback & kill-switch

SLOs & Observability

Telemetry
  • • OpenTelemetry tracing & metrics
  • • Prometheus/Grafana dashboards
  • • Structured logs (Loki/ELK)
DB & LLM Evals
  • • pg_stat_statements & pgbouncer stats
  • • Eval suites (precision/recall, hallucination)
  • • Cost & token budget tracking
SLOs
  • • API P95/P99 latency targets
  • • Error budget & reliability
  • • Rate-limit & autoscaling policies

Industries & Use-Cases

GovTech / LegalTech Fintech & Risk Ops & Analytics E-commerce Geo / 3D Visualization

Let’s build something great

Tell us about your goals. We’ll reply with ideas and a clear next step.

Contact

  • 📧 [email protected]
  • 🌐 Indonesia || Singapore || Hongkong
  • 🔒 NDA-friendly, security-first
We respect your time and privacy. Zero spam.