Open to new work

Full stack AI engineer shipping production systems.

5+ years shipping AI products across image, text, and video. Healthcare and fintech platforms in production. Currently leading backend at Centrox AI, building agents, retrieval pipelines, computer vision, and the cloud infra to run them.

Centrox AI
Islamabad / Remote
Full Stack / Cloud / AI
Open to Remote
Freelance & Client Work
Consultant
// who's behind this
const me = {
  name: "Adeel Irshad",
  role: "Full Stack Engineer",
  based_in: "Islamabad, PK",
  works_at: "Centrox AI",
  stack: [
    "TypeScript", "Python", 
    "Go", "NestJS", 
    "FastAPI", "Next.js", 
    "MongoDB", "Kubernetes", 
    "OpenAI", "LangChain"
  ],
  open_to: "remote, freelance, consulting"
};

Production stack across backend, AI, and infra.

The stack I use to build, ship, and run production software.

Backend & APIs

Node.js, NestJS, Express, Python, FastAPI, Go, REST APIs, GraphQL, WebSockets, BullMQ, Celery

Frontend & Product UI

Next.js, React, TypeScript, Tailwind CSS, App Router, Server Components, SSR, dashboards and landing pages

Cloud, Kubernetes & Delivery

Kubernetes, K3s, Docker, Helm, Hetzner, GitLab CI, Nginx, Cloudflare, Linux ops and production deployments

Data, Caching & Search

PostgreSQL, MongoDB, Redis, Mongoose, data modeling, background queues, real time pipelines and API integrations

LLMs, Agents & RAG

LLM APIs, agent workflows, retrieval pipelines, prompt orchestration, eval driven iteration, and automation systems

AI Vision & OCR

Vision models, OpenCV, OCR pipelines, document forensics, tampering detection, MRZ validation, face matching, and fraud scoring

Integrations & Platform Ops

Stripe, Twilio, Keycloak, Brevo, Backblaze B2, webhook flows, cron automation, observability and support for production incidents

What I've built.

AI products, backend platforms, fintech systems, and the infrastructure behind them.

How they were built.

Architecture, tradeoffs, and the outcomes each system had to support.

ScriptLix project preview
Stack
GoFiberPostgreSQLRedisFirebaseOpenAIStripeGoogle Cloud Storage+5
Architecture
  • Multi-stage LLM orchestration: outline to beats to scenes to dialogue, each with context handoff
  • Structural validation pass (act breaks, page counts, character consistency) gates persistence
  • Versioned screenplay store so users can iterate without losing prior drafts
  • Real-time style customization (director, tone, location) without re-generating from scratch
AI

AI screenplay generation: from one-line idea to feature-length script.

ScriptLix turns a concept into a directed, formatted screenplay. The pipeline orchestrates LLM calls for outline, beats, scenes, and dialogue, then validates structure (act breaks, page count, character consistency) before persisting to a versioned screenplay store. Real-time customization (director style, tone, location), script-health analysis, and a personal screenplay library.

Multi-stage
LLM Pipeline
Real-time
Customization
Versioned
Drafts
Zygenai project preview
Stack
PythonFastAPIPostgreSQLStripeOpenAIeBay APINext.js
Architecture
  • eBay OAuth and token refresh per seller, isolated workspace per account
  • LLM-driven message classifier and draft-reply generator with brand-tone retrieval
  • Stripe subscription billing with Stripe Customer Portal for upgrades and downgrades
  • Audit log for every agent action (draft, send, suppress) for trust and rollback
AI Agent

An AI agent that runs the long-tail of an eBay seller's day.

Zygenai automates listing optimization, customer messaging, and order workflows for eBay sellers. eBay OAuth onboarding, Stripe subscription billing, multi-seller dashboard. The agent watches incoming messages, drafts replies in the seller's tone, and queues actions for human approval.

Multi-seller
SaaS
OAuth
eBay Connect
Audited
Agent Actions
Sarmaaya.pk project preview
Stack
PythonApache AirflowMongoDBDjangoStripeCelery
Architecture
  • Apache Airflow DAGs orchestrating per-asset ingestion (stocks, crypto, forex, commodities)
  • Source-agnostic schema normalization into a single MongoDB store
  • Cached read-paths for sub-second query latency at 100K+ unique visitor scale
  • Django savings platform with Stripe Connect for community committee payouts
Fintech

Real-Time Multi-Asset Data Pipelines for 20K Monthly Users

Built Apache Airflow pipelines in Python pulling and unifying live stock, crypto, forex, and commodity feeds from multiple external sources into a single normalized MongoDB store. Plus a Django savings platform with Stripe payment workflows for community committees.

100K+
Unique Visitors
20K
Monthly Users
Airflow
Pipeline Orchestration
Bux project preview
Stack
PythonFastAPIUvicornLangChainOpenAIPineconeMongoDBMotor+3
Architecture
  • Transcription and chunking pipeline per episode, with metadata for guest, episode, timestamps
  • Vector indexing in Pinecone for semantic retrieval across the podcast corpus
  • LangChain RAG flow with grounded answers and citation back to specific episodes
  • Next.js chat UI with streaming responses and per-episode drill-down
AI / RAG

An AI companion for the Beyond UX Design podcast.

Bux ingests every podcast episode, indexes it into a vector store, and lets fans chat with the show's content. Conversations are grounded in retrieved transcript chunks. Out-of-corpus questions get gracefully refused.

Pinecone
Vector Store
RAG
Grounded Answers
Multi-episode
Corpus
ClearSky Co project preview
Stack
NestJSTypeScriptMongoDBFirebaseHetznerDockerNext.jsReact+3
Architecture
  • NestJS service with TypeScript across modules (auth, jobs, billing, integrations)
  • MongoDB primary store; Firebase for auth and realtime triggers
  • Dockerized deploys on Hetzner with REST API contract for the frontend
  • Next.js admin UI with React Hook Form, Zod, and Radix UI primitives
Client Work

Production backend and DevOps for an Australian SaaS client.

End-to-end build for ClearSky Co. NestJS service backed by MongoDB, Firebase auth, Dockerized deploys on Hetzner, and a Next.js admin frontend with Radix UI and Zod-validated forms. Live in production.

NestJS
Backend
Hetzner
Deployed
Live
In Production

Get in touch.

Open to AI engineering, backend, and full stack work. Remote, freelance, or consulting.

Start a conversation