ABRAR AHMAD — AI/ML & CLOUD ENGINEER
Building IntelligentSystems That Think,Learn & Scale.
AI/ML Engineer specializing in Generative AI, Agentic AI, RAG, Computer Vision, MLOps and AWS Cloud Engineering.
Open to AI/ML & Cloud roles and freelance workEngineering intelligence, end to end.
I'm Abrar Ahmad — an AI/ML Engineer and AWS Cloud Engineer who bridges the critical divide between experimental machine learning models and scalable, production-grade cloud software.
My expertise centers around the frontier of Generative AI (autonomous multi-agent workflows, hybrid RAG with vector retrieval, and contextual reasoning) paired with real-time Computer Vision (detection, tracking, and automated safety systems using YOLOv8 and PyTorch).
Rather than isolated demos, I build complete systems: from data pipelines and model optimization to containerized FastAPI microservices and fault-tolerant cloud infrastructure on AWS.
Generative & Agentic AI
Enterprise RAG pipelines, autonomous multi-agent teams, and LLM applications that reason over proprietary knowledge bases with deterministic tool-calling precision.
Computer Vision & Safety
Real-time detection, tracking, license plate recognition, and industrial safety systems engineered with YOLOv8, OpenCV, and PyTorch for edge and cloud deployment.
High-Performance ML Backends
Production-ready async Python and FastAPI microservices that serve models with ultra-low latency, strict response validation, and modular architecture.
MLOps & AWS Cloud Engineering
Deploying, containerizing with Docker, and scaling intelligent systems on AWS (ECS, Lambda, S3, EC2) backed by automated CI/CD and observability.
How I architect intelligent systems.
Core principles guiding every system from initial model formulation to production rollout.
Production Over Prototypes
Moving beyond quick notebooks into containerized, low-latency microservices engineered for high concurrency and real-world reliability.
Autonomy With Strict Guardrails
Architecting agents and RAG workflows with structured output validation, hallucination checks, and reliable fallback loops.
Cloud-Native & Cost-Optimized
Designing resilient AWS infrastructure that scales horizontally under load while actively controlling token consumption and inference costs.
Five domains that combine into systems which perceive, reason and scale in production.
Generative AI
Building applications powered by large language models and generative systems.
Agentic AI
Designing autonomous agents that reason, plan and use tools to complete tasks.
RAG
Retrieval-Augmented Generation grounding LLM answers in a knowledge base.
LangChain
Orchestrating LLM chains, retrievers and tools into production pipelines.
LLM Applications
End-to-end apps built on top of hosted and open large language models.
AI Agents
Multi-step agents that combine reasoning with external APIs and memory.
Nexus Bids AI — Autonomous Multi-Agent RFP Bidder
An enterprise-grade autonomous multi-agent AI system that parses RFPs, evaluates Go/No-Go feasibility, retrieves institutional knowledge via ChromaDB RAG, and generates production-ready technical proposals and cost estimates.
Concrete Cracks Detection
Deep learning segmentation model that highlights crack regions at pixel level for faster, safer inspection of bridges, roads and buildings.
Enterprise Digital Twin — AI CEO Simulator
A multi-agent system where an AI CEO delegates to Marketing, Finance and HR agents, runs SimPy simulations before every decision, and stays grounded in company policy documents via vector memory.
Office Activity Tracking System
Real-time workplace analytics that classifies activities such as working, roaming, sleeping and talking, then reports working vs non-working time.
VeriFact AI — Multi-Agent Fake News Detector
A CrewAI agent team that fact-checks news claims against live web and Wikipedia sources and scores them from 0 to 100, served from a Dockerized FastAPI backend.
LinkTrim — URL Shortener & Real-Time Analytics
A production-style URL shortener with Snowflake IDs, geo-routing, expiring links, asynchronous click logging and a ClickHouse-powered analytics dashboard.
From raw data to deployed intelligence.
Every system I ship follows a disciplined, production-grade path — combining modern AI models, rigorous evaluation, and resilient AWS cloud infrastructure.
Autonomous multi-agent orchestration, hybrid semantic retrieval, and tool-augmented generation with zero-hallucination verification loops.
COGNITIVE ARCHITECTUREIngest & Chunk
Parse unstructured PDFs, tables, and docs with semantic chunking, extracting high-dimensional embeddings into vector databases.
Hybrid Retrieval
Execute two-stage hybrid retrieval combining dense semantic similarity with BM25 keyword matching and cross-encoder re-ranking.
Agent Reasoning
LangGraph orchestrates autonomous sub-agents with dynamic tool calling, structured Pydantic schemas, and citation verification.
FastAPI & AWS Serve
Expose high-throughput streaming endpoints with async FastAPI, containerized in Docker and autoscaled across AWS ECS Fargate.
9 min readTrain YOLOv8 on a Custom Dataset in Google Colab
Step-by-step guide to training YOLOv8 on your own dataset in Google Colab: dataset format, data.yaml, training, metrics, inference and export.
Read post
11 min readYOLOv8 Metrics Explained: Precision, Recall and mAP
What precision, recall, mAP50 and mAP50-95 really mean in YOLOv8, explained with real training results from four of my models, and how to act on them.
Read post
Open to AI/ML engineering, Cloud Computing roles, freelance work and collaborations on ambitious systems. Reach out through any channel below.