Hasan Iqbal Butt — MLOps & RAG Platform Engineer. I build the infrastructure that takes AI from prototype to production across 18+ production systems. Enable JavaScript for the interactive datacenter and retrieval pipeline visualization.
Filter by category
An institute running on paper registers, digitalized end-to-end -- bilingual Urdu/English, live with 200+ users.
Hybrid AWS + GCP infrastructure for a BNPL fintech -- 17 microservices, PCI DSS scope, bank VPN integrations.
An EKS platform with real SRE discipline -- SLOs, Karpenter, and 99.9% measured over six months.
Six hand-configured servers to a reproducible K3s + Postgres HA stack -- full rebuild in under 30 minutes.
Fargate pricing said $8k/month. ECS on EC2 with Reserved Instances said $3.8k. High-availability booking at scale for less.
Productizing a legal RAG pipeline: auth, rate limits, and 99.95% uptime over six months.
GDPR erasure across 12 systems in 72 hours, with Step Functions execution logs as audit evidence.
14 services, 3 clusters, zero drift -- 120 deploys/week through ArgoCD with auto-rollback.
OCR pipeline for degraded handwritten manuscripts -- preprocessing did the heavy lifting, not the model.
An 18GB transformer on serverless: cold start cut from 2+ minutes to 8 seconds with gcsfuse streaming.
A model that worked on laptops and died in prod -- rebuilt as an autoscaled SageMaker endpoint, <200ms P99.
200 attorneys, $18k/month in API calls -- moved to self-hosted A100s at 82% lower cost.
Replaced an $800/mo Pinecone instance with pgvector -- measured on a domain golden set, not vibes.
Legal RAG that respects document structure: 5 retrieval layers over 3,145 docs, each layer fixing a real failure.
214 undocumented ERP tables turned into 42 curated MCP tools an agent can actually use safely.
A JS-rendered council portal with session pagination, scraped responsibly -- zero missed applications.