AI agent engineer, full stack developer, DevOps and cloud
I build the parts that keep production AI agents safe, affordable and online: spend limits, prompt injection defence, audit trails, Kubernetes deployments, CI/CD and load tested infrastructure on AWS and Docker. I also build the full stack product around them, from Angular and React frontends to Node.js backends.
Agents that survive production, and the full stack around them
I am a full stack developer who has built web products since 2022 with Angular, React, Node.js and Python. Since 2024 my focus has been AI agents, including voice and RAG, and the security, cost and infrastructure around them. I run apps on AWS and ship them with Docker and CI/CD.
My strongest work is where an agent meets the real world: what happens when a tool call hangs, a prompt carries an attack, a bill spikes or a database goes down. I measure those cases, fix what breaks and publish the numbers, including the weak ones. I am based in PKT (UTC+5) and work remotely.
Skills
What I work with
AI agents
LLM agents, RAG, embeddings, vector search, voice agents with Pipecat and LiveKit, LangChain, Ollama
Security
Prompt injection detection (DeBERTa classifier), per request token and budget caps, append only audit logging, AWS IAM and security groups
DevOps and cloud
AWS (EC2, Auto Scaling, ELB, CloudFront, RDS, S3, Lambda, API Gateway, IAM, security groups, Bedrock, SageMaker), CI/CD with GitHub Actions, Docker and Docker Compose, Kubernetes, Helm, Nginx
Built a spend guard in front of LLM APIs that caps tokens and budget per request. A capped request returned 64 output tokens instead of 2,618.
Packaged it as a Docker Compose stack and wrote a Helm chart for Kubernetes. A rolling restart of 6 pods had 0 failed requests out of 287,403.
Built a voice agent on Pipecat and LiveKit that handles several calls at the same time, and ran apps on AWS with load balancers, CloudFront, IAM and cost management.
Full Stack Engineer, IT Pixelz
Mar 2023 to Jun 2024
Built booking flows for a doctor booking platform serving about 50 doctors across 20 clinics, using React, Node.js, Express and MongoDB.
Built Angular 17 features for an enterprise task management app and cut the main bundle from 6 MB to 2 MB.
Shipped Node.js and .NET microservices on Redis, MySQL and MongoDB, deployed to AWS with Docker and CI/CD.
Angular Developer (part time), Zepto Systems Limited
Jan 2022 to Feb 2023
Built the frontend of an IT staffing comparison platform, later also backend and CMS work.
Built a shared design system and component library with Angular Material and Bootstrap.
Improved load times with lazy loading, code splitting and fewer API calls.
Selected work
BleedStop: runtime cost and security guard for AI agents
A control plane that sits in front of LLM, SMS and payment APIs. Apps ask before they spend, and get back ALLOW, LIMIT, CHALLENGE or DENY from deterministic rules.
5,433/sdecisions per second, p99 36.2 ms, 0 failed of 60,000 (Compose stack, one 8 vCPU machine)
0 of 287,403requests failed during a 6 pod Kubernetes rolling restart
97.6%fewer output tokens on a capped request (64 instead of 2,618)
77.5%of held out injections caught at a 5% false positive rate
Agent Receptionist: a tool-calling agent I can measure
An LLM agent for appointment businesses with cited RAG answers, rule-based handling of emergencies and opening hours at zero tokens, and a per-sentence guard. Deterministic eval 44/44, live Groq eval 43/44, live 20-question test 19/20, on fictional hotel and dental businesses. Source is private; available on request.
73 second walkthrough with voiceover: case study, live demo and the same engine on a second business.
Built a voice agent that handles several calls at the same time, works out what the caller needs and passes the details to staff on WhatsApp.
RAG retrieval and agent pipelines
Built retrieval with embeddings and vector search, plus an agent pipeline that turns a website URL into a finished video: site scan, script, Remotion templates, stock media and export.
DevOps and cloud work
AWS
Run apps on AWS with EC2 behind load balancers, CloudFront, tightened security groups and IAM, and cost management. Earlier work deployed to EC2, Auto Scaling, RDS, S3, Lambda and API Gateway.
Docker and CI/CD
Packaged BleedStop as a Docker Compose stack with 3 API replicas, Redis, Postgres and nginx. Set up GitHub Actions CI for it, and shipped services to AWS with Docker and CI/CD.
Kubernetes and Helm
Wrote a Helm chart with readiness and liveness probes, a PodDisruptionBudget, HPA and graceful drain. A rolling restart of 6 pods had 0 failed requests out of 287,403.
Full stack work
Doctor booking platform
Built the booking flows with React, Node.js, Express and MongoDB for about 50 doctors across 20 clinics. Included Gemini assisted booking, live updates over WebSockets, reminders and JWT auth.
Enterprise task management app
Built Angular 17 features with RxJS, NGXS and Reactive Forms. Cut the main bundle from 6 MB to 2 MB with lazy loading and Tailwind.
Microservices and dashboards
Shipped Node.js and .NET microservices on Redis, MySQL and MongoDB, deployed to AWS with Docker and CI/CD. Built dashboards and reporting views in a CRM used to manage hundreds of client accounts.
Open source
Fixes in the frameworks agents run on
Merged
Pipecat #5760
Function call timeouts are no longer cancelled by intermediate results. Added tests.
I am open to full time remote roles and contracts. If you are putting AI agents into production, or need a full stack engineer who cares about security, cost and reliability, send me a note.