Hamza Shaikh

Site reliability and infrastructure engineer.
I keep production systems observable, quick to recover, and cheap to run.

In practice that's validator telemetry, Prometheus alerting, CI/CD, and the production cloud underneath them, usually with me as the only person on call.

Werkstudent, internship and junior roles in Germany140 full / 280 half days work authorisation. Completed B.E. is ANABIN-verified, so EU Blue Card eligibility does not depend on finishing the M.Sc.

Kaiserslautern, DE · Dammam, KSA · Mumbai, IND

AWARD · 2026

Winner · Superteam Germany Solana Ideathon

LIVE SYSTEM

SentinelSOL · Solana validator observability

M.SC.

Distributed Systems · RPTU Kaiserslautern

PRODUCTION

Sole DevOps owner · −27% AWS spend · $230→$168/mo

Selected Projects

DAEMONlive

SentinelSOL

Separate-process telemetry and anomaly detection for Solana validators.

What I built The Go telemetry daemon and the 3σ alert rules. I also fixed the denominator bug that silenced the alert as a validator stalled.

Winner, Superteam Germany / neosfer Ideathon Frankfurt 2026.

Live SystemSource Code

A Go daemon polls the validator's Solana JSON-RPC out of process and exports vote-credit accrual and slot progression to Prometheus. 3-sigma Z-score over a rolling one-hour baseline, routed through Alertmanager to Telegram.

SentinelSOL Telemetry
SNAPSHOT
3σZ-score anomaly window10spoll + scrape interval5srefresh cadence
Architecture Overview
PIPELINE
Solana JSON-RPC
Go telemetry daemon
Prometheus / PromQL
Alertmanager → Telegram
Incident Replay
NOMINALT+00:00
  1. T+00:00
    Vote Credit VelocityBASELINE

    Per-slot vote-credit accrual tracked against a rolling one-hour mean. Nominal.

  2. T+00:06
    Deviation DetectedDETECT

    Accrual rate drifts below the mean; earnings slow before a single vote is missed.

  3. T+00:12
    Threshold BreachedDETECT

    Z-score crosses 3σ against the rolling baseline. Noise filtered out; this is real.

  4. T+01:12
    PromQL EvaluationDETECT

    Recording rule feeds the alert expression; it stays true across the for: 1m window.

  5. T+01:13
    Alertmanager TriggeredALERT

    Grouped, deduplicated, routed by severity. One page, no flapping.

  6. T+01:14
    Telegram NotificationALERT

    Operator paged in-channel with validator identity, the metric, and its current value.

  7. T+02:00
    Operator InvestigationRESPOND

    Check peer set and slot progression; correlate with the epoch boundary.

  8. T+06:30
    ResolutionRESOLVED

    Peer set reconnected before delinquency. Credit velocity re-stabilises; 0 slots lost.

1m13sdetect to page · automated~5mpage to resolve · operator0slots delinquent
GoPrometheusPromQLAlertmanagerTelegram APIDocker Compose
NODEin development

AutoSRE

Chaos engineering platform: inject failure into live services and prove the resilience patterns hold.

What I built End to end: the fault-injection control plane, the circuit breaker and rate limiter, RED metrics with SLO alert rules, and the security-scanned CI.

Live SystemSource Code

Injects latency, errors, CPU burn and memory pressure into live services through a REST control plane. Experiments are bounded, time-limited and self-reverting. RED metrics via ASGI middleware against a p95 500ms SLO, alerting at 750ms.

AutoSRE · Architecture
DATA FLOW · Metric
  1. Metric

    RED metrics on every request through ASGI middleware, against a p95 500ms SLO.

  2. Alert

    The budget breaches; Alertmanager fires, grouped and deduplicated by severity.

  3. Bound

    Every experiment is capped at two layers, time-limited and self-reverting; the control plane is exempt from injection.

  4. Recovery

    Per-client token-bucket rate limiting returns 429 with Retry-After; a three-state circuit breaker sheds load from a failing dependency.

  5. Healthy

    Readiness returns 503 while a dependency's breaker is open. Auto-remediation is deliberately out of scope.

500msp95 latency SLOREDrate · errors · durationCI/CDGitHub Actions gated
PythonFastAPIPrometheusDockerDocker ComposeGitHub Actions
VECTORshipped

GridCast

LSTM-based regional temperature forecasting system.

IEEE ICSPCRE 2024 · Paper ID 652 (submitted)

Live System

28 LSTM models across 7 regions × 4 IMD seasons. 70 years of daily max temperature grids (1951–2021). 14-step input window, 7-step forecast. Optuna TPE per-subset hyperparameter search.

GridCast · Architecture
DATA FLOW · Dataset
  1. Dataset

    70 years of IMD daily-max temperature grids, 1951 to 2021, partitioned by region.

  2. Training

    28 LSTM models, one per region and IMD season, tuned with Optuna TPE per subset.

  3. Forecast

    A 14-step input window predicts a 7-day temperature horizon per cell.

  4. Output

    A Flask inference API serves a React and D3 heatwave map.

28region × season models70 yrsdaily temp grids7-dayforecast horizon
PythonTensorFlowKerasLSTMFlaskReact.jsOptunaGeoPandas

Experience

Patil Kaki· Shark Tank India D2C food brand

DevOps Engineering Intern, sole infrastructure owner

Jun – Jul 2023

Sole infrastructure engineer in a 10-person team, behind a storefront running roughly 2,500 orders a month.

  • Migrated production from EC2 to ECS/Fargate; dynamic task provisioning cut AWS spend 27%, $230 to $168 a month
  • Root-caused a three-day chatbot outage to a whitespace character intermittently injected into the integration URL
  • Built the AWS network and access layer: VPC subnets, routing, security groups, S3, and least-privilege IAM scoped separately for founders, developers and logistics staff
  • GitHub Actions pipelines gating main-branch deploys with rollback, supporting weekly releases
  • Moved fulfilment off the synchronous checkout path with a BullMQ-on-Redis queue

CMP Infotech· Microsoft Partner

Java Development Intern

Dec 2021 – Jan 2022

Desktop application development in Java.

  • School management system with multi-user access control
  • Java Swing UI with MySQL persistence via JDBC

Education

RPTU Kaiserslautern-Landau· Germany

M.Sc. Computer Science, Distributed Systems · Software Engineering

Oct 2025 – Expected 2028

Career Break & M.Sc. Admissions· Dammam

Graduate applications and German student visa process, Intake deferred to Oct 2025

Jul 2024 – Sep 2025

Xavier Institute of Engineering· Mumbai

B.E. Information Technology, CGPA 8.69/10 (German equivalent 1.6)

Aug 2020 – Jun 2024

Skills

What I reach for, grouped by where it sits in the stack. Listed because I've shipped or operated something with it, not to pad a list.

Infrastructure

DockerDocker ComposeLinuxNGINXKubernetes (foundational)

Observability

PrometheusPromQLGrafanaAlertmanagerRED metricsSLOs

Cloud

AWSCloudflareDigitalOcean

Languages

GoPythonBashC++

Automation

GitGitHub ActionsCI/CD

Security

TrivyBanditpip-auditgovulncheck

Storage & APIs

RedisBullMQMySQLFastAPI

Testing

pytestchaos engineering

Certifications

AWS Certified Cloud Practitioner2023
Postman API Fundamentals · Student Expert2023

Curriculum Vitae

PERSONNEL FILE

Hamza Shaikh

Infrastructure and reliability engineer. Sole DevOps owner at a Shark Tank India D2C food brand, now doing a distributed systems master's at RPTU. Shipped SentinelSOL, building AutoSRE.

Download PDF

Last updated · 6 NOV 2026

RoleSite Reliability · DevOps · Cloud Infrastructure
FocusObservability, failure testing, cloud cost
EducationM.Sc. Computer Science · RPTU Kaiserslautern
BaseKaiserslautern, Germany
Open toWerkstudent, internship and junior roles in Germany · 140 full / 280 half days work authorisation. Completed B.E. is ANABIN-verified, so EU Blue Card eligibility does not depend on finishing the M.Sc.

CAPABILITIES

Sole DevOps ownership of a production stack
Prometheus / PromQL / Alertmanager observability
AWS ECS / Fargate and Docker Compose operations
GitHub Actions CI/CD gating main-branch deploys with rollback
Python and Go automation
Hamza_Shaikh_CV.pdf
Open PDF

Contact

Open to Werkstudent, internship and junior roles in Germany · 140 full / 280 half days work authorisation. Completed B.E. is ANABIN-verified, so EU Blue Card eligibility does not depend on finishing the M.Sc.Fastest reach: Telegram or email.