Ikram Khan · Full-Stack AI Engineer
AI Engineer.
RAG, LLMs & Agents, shipped.
I build, deploy, and scale AI-powered applications for startups and SaaS teams.
Top Rated Upwork freelancer specializing in production LLM applications: Retrieval-Augmented Generation (RAG) systems, LLM fine-tuning, AI agents with LangChain and LangGraph, and full-stack AI apps on AWS with Python, FastAPI, Rust, and React.
Upwork Track Record
Consistently exceeding client expectations with high-quality, scalable code.
What I Can Build.
AI agents, RAG systems, LLM fine-tuning, and full-stack AI apps, from architecture to production deployment on AWS.
AI Agents & RAG Systems
Tool-using AI agents and production RAG pipelines built with LangChain and LangGraph: hybrid search, reranking, and evaluation so your assistant answers accurately from your own documents and knowledge base.
Full-Stack AI Apps
End-to-end product builds pairing high-performance AI backends (FastAPI, Axum, Django) with responsive React frontends: clean APIs, streaming LLM responses, and secure multi-tenant data isolation.
LLM Fine-Tuning & MLOps
Fine-tune open-source LLMs and vision models (LLaMA 3, Hugging Face) on your data, then deploy on AWS SageMaker, Bedrock, or Hugging Face endpoints with Docker and Terraform.
The Toolkit
- Python
- TypeScript
- JavaScript
- Rust
- FastAPI
- Django
- React.js
- Axum
- LangChain
- OpenAI
- HuggingFace
- Ollama
- Candle.rs
- AWS Cloud
- Docker
- Terraform
- CI/CD
- Kubernetes
Selected AI Projects.
Production RAG, AI agent, computer vision, and MLOps applications solving real-world problems.
Khan Education: AI-Powered E-Learning Platform
Open-source e-learning platform with interactive lessons, quizzes, and powerful admin tooling — delivered globally through a serverless AWS edge architecture.
The Challenge
Building a scalable, low-latency educational platform that handles dynamic content delivery and complex user administration globally.
The Solution
Implemented a serverless architecture using AWS CloudFront for edge caching and S3 for static assets, managed via Terraform for reproducible infrastructure.
The Outcome
Achieved sub-100ms load times for global users and reduced infrastructure costs by 40% compared to traditional EC2 hosting.
Docgram: Chat-with-PDF RAG Social Platform
AI-powered social platform where users share PDFs in an Instagram-style feed — every document becomes conversational through integrated AI chat built on Gemini and Pinecone.
The Challenge
Enabling meaningful social interaction around static PDF documents while providing instant, context-aware answers to user queries.
The Solution
Developed a RAG pipeline using Pinecone for vector storage and Gemini for generation, integrated into a familiar social feed UI.
The Outcome
Created a unique 'Chat with PDF' social experience, processing thousands of document pages with high retrieval accuracy.
Mediscribe: AI Medical Transcription & Clinical Summaries
Healthcare proof of concept that transcribes patient consultations and generates clinical summaries using Gemini's audio and language capabilities.
The Challenge
Accurately transcribing and summarizing complex medical terminology in real-time while ensuring data security.
The Solution
Leveraged Gemini's multimodal capabilities for high-fidelity transcription and PynamoDB for secure metadata handling on AWS.
The Outcome
Delivered a POC demonstrating 95% accuracy in medical term recognition and automated summarization for EMR entry.
Health AI Agent System: LangGraph Multi-Agent Workflow
Multi-agent system built on LangGraph that automates wellness checks, personalized video recommendations, and document Q&A.
The Challenge
Orchestrating multiple specialized AI agents (Wellness, Video, Classes) to provide a cohesive user experience without hallucinations.
The Solution
Built a stateful multi-agent system using LangGraph to route requests intelligently between vector DBs and external APIs.
The Outcome
Streamlined patient engagement workflows, reducing manual wellness check calls by automating initial triage.
Passport MRZ Extraction & Face Verification
Fast, secure identity-verification app that extracts passport data via MRZ parsing and confirms identity by face-matching against a selfie.
The Challenge
Automating identity verification with high reliability to prevent fraud in a high-stakes environment.
The Solution
Integrated MRZ parsing libraries with custom face-matching algorithms and deployed on a secure AWS environment.
The Outcome
Reduced identity verification time from minutes to seconds with a high confidence score for fraud detection.
Retail Demand Forecasting with XGBoost on SageMaker
Automated demand-forecasting pipeline built with XGBoost on AWS that optimizes retail inventory — cutting carrying costs while sharply improving forecast accuracy.
The Challenge
A retail chain was losing $5M annually to overstock and stockouts, with manual forecasting taking 3 days per product.
The Solution
Created a self-service XGBoost pipeline on AWS that auto-generates optimized sell-through forecasts for any product.
The Outcome
Improved forecast accuracy to 92%, reduced inventory costs by 15%, and cut forecasting time from 6 hours to under 15 minutes.
RoBERTa Fake News Detection: SageMaker MLOps Pipeline
End-to-end AWS SageMaker pipeline for fake-news detection — automated preprocessing, RoBERTa training, evaluation, and deployment.
The Challenge
Creating a reproducible, end-to-end ML pipeline that automatically retrains and deploys models as new data arrives.
The Solution
Architected a SageMaker pipeline covering preprocessing to deployment, using RoBERTa for state-of-the-art NLP performance.
The Outcome
Established a robust MLOps workflow, ensuring the fake news detection model remains current with evolving misinformation trends.
Experience & Certifications.
A timeline of consistent delivery, continuous learning, and professional excellence.
Employment
Freelance AI/ML Engineer
Apr 2024 - Present18 client projects on RAG systems, LLM fine-tuning, AI agents, and chatbot development. Top Rated with 100% Job Success and a 5.0/5.0 rating.
Data Analyst
Jan 2022 - Mar 2024Cleaned and analyzed large datasets with Python and built predictive models to improve customer satisfaction.
Education & Certs
Bachelor of Science in Software Engineering
Balochistan University of IT and Management Sciences
2019 - 2022"Working with Ikram was 10/10 great once again. Looking forward to continuing to work together!"
John Meyer @ Freshprint
RAG Compliance System