shahryar khalid - AI Agent Developer | ContraWork by shahryar khalid
shahryar khalid

shahryar khalid

I am AI engineer with specialized in RAG system and ai agent

New to Contra

shahryar is building their profile!

Followed by GALLERY L

Lahore, Pakistan

WorkPostsAbout

Lahore, Pakistan

Cover image for A conversational AI job search
A conversational AI job search assistant that finds relevant, up-to-date job listings from the web based on what you ask for, instead of making you scroll through job boards. Users describe the role they want in plain language, and the agent searches live sources and returns matching opportunities in a chat interface. It is built as a multi-agent system on LangGraph, with specialized agents coordinating the search. Live web data comes through the Tavily MCP (Model Context Protocol) integration, and the language model is served by Groq for fast responses. Conversation state is saved with SQLite checkpointing, so the agent remembers context across a session. The FastAPI backend streams responses in real time over Server-Sent Events (SSE), and the frontend is built with HTML, CSS, and JavaScript. Tech stack: LangGraph, Tavily MCP, Groq, FastAPI (SSE streaming), SQLite, HTML/CSS/JS Source code: github.com/shahryarkhalid-cmd/Job-seeking-agent I build agents like this for businesses that need automated research, lead generation, or data gathering.
0
1
QanoonAI is an AI-powered legal assistant for Pakistani law. Users ask legal questions in English or Urdu and get clear, grounded answers drawn from Pakistan’s legal documents, with conversation history kept across sessions. It uses a Corrective RAG (CRAG) pipeline: relevant legal text is retrieved from a FAISS vector store and re-ranked with a cross-encoder for precision. If the retrieved context isn’t good enough, the system falls back to live web search through Tavily instead of guessing. This reduces hallucinations, which matters a lot in legal use cases. Tech stack: LangGraph (agent orchestration, SQLite checkpointing for memory), FastAPI with streaming (SSE) responses, Streamlit frontend, PyMuPDF, LangSmith (tracing and debugging), and Docker (distroless image) for deployment. I can build similar document-grounded assistants for law firms, clinics, universities, and businesses that need reliable answers from their own documents.
0
8
AI engineer portfolio with a built-in AI assistant, “Onyx,” that answers visitor questions about my work. Built with React, Tailwind, and Framer Motion on the frontend and LangGraph, FastAPI, and Groq on the backend. Deployed on Azure.
1
10
:Tixora — Event Ticketing Platform with Fraud Detection A production-grade event ticketing system built end-to-end to handle real-world booking traffic, payments, and fraud — not just a CRUD demo. What it does: Tixora lets users browse events, book tickets, and manage bookings — while the backend handles the hard parts: preventing overselling under concurrent demand, processing payments reliably, and catching fraudulent activity before it becomes a problem. Key engineering decisions: ⚙️ FastAPI + PostgreSQL (Supabase) for a scalable, type-safe backend 🔒 Redis distributed locking to guarantee no double-booking or overselling during high-traffic ticket drops 💳 Stripe integration with webhooks for real-time, reliable payment confirmation ⏱️ APScheduler to auto-expire abandoned/pending orders and free up inventory 🛡️ ML-based fraud detection (XGBoost/scikit-learn) to flag suspicious booking patterns 🎫 QR-code ticket generation for fast, secure check-in 🐳 Dockerized deployment on Railway Why it matters: Ticketing systems fail in production in specific, predictable ways — race conditions on popular events, payment/webhook mismatches, silent fraud. Tixora is built to handle these from day one rather than patch them after launch. 🎥 Demo: https://youtu.be/jhe2qY3ZdL8 (https://youtu.be/jhe2qY3ZdL8)💻 Code: https://github.com/shahryarkhalid-cmd/Event-Ticketing-Platform-with-Fraud-Detection Available for backend/API development, RAG systems, and agentic AI projects.
1
28