Rehan Malik rehan243

AI / ML Engineer - LLM Systems - RAG - Multi-Agent - RLHF - MLOps
_{building production AI that ships to real users, not demos}

about me

i'm an AI/ML engineer. right now i'm building production AI systems at Reallytics.ai and Verticiti, mostly getting large language models to do useful things in the real world. not demos, actual systems with real users and real traffic.

before this i was at Afiniti and Cloud Kinetics for a few years. fraud detection, voice analytics, enterprise search. the kind of stuff that pages you at 3am when something breaks.

honestly what keeps me going is when an agent you built solves something you never explicitly told it to do. that feeling never gets old.

what i'm working on right now:

multi-agent systems that don't fall apart when you chain them
RAG pipelines that actually return relevant results
writing about what i learn every day, check it out here

featured projects

Agentic AI Workflows 8 specialized AI agents with LangChain + OpenAI function calling. multi-agent orchestration with planning loops and guardrails. the project i'm most excited about right now.	RAG Enterprise Search production-style retrieval pipeline: hybrid dense+sparse search with FAISS and BM25, cross-encoder re-ranking, built to scale on AWS SageMaker.
Voice AI Platform real-time voice infrastructure built for high-concurrency calls. WebSockets, Kafka, VAD, streaming STT, with a from-scratch sentiment analysis component.	LLM Fine-Tuning LoRA fine-tuning LLaMA and Mistral with LoRA/QLoRA/PEFT, a cost-efficient alternative to hosted APIs. includes the full training loop, data pipeline, merge + quantize scripts.
RLHF LLM Optimization full RLHF pipeline: reward model with Bradley-Terry loss, PPO trainer with KL scheduling, DPO as an alternative, plus an eval harness and safety-compliance checks.	Sentinel Fraud Detection ensemble XGBoost + neural net with a rich engineered feature set. Redis-backed real-time velocity scoring, SHAP explainability, Kafka alert routing.

tech stack

not going to pretend i use everything equally. here's what i actually reach for:

the full picture (click to expand)


daily drivers	Python, PyTorch, FastAPI, Docker, Git, VS Code
LLM and GenAI	LangChain, LlamaIndex, HuggingFace Transformers, vLLM, PEFT/LoRA/QLoRA
data and vector	FAISS, ChromaDB, Pinecone, PostgreSQL, MongoDB, Redis, Kafka, Elasticsearch
cloud and MLOps	AWS (SageMaker, Bedrock, Lambda, ECS), GCP Vertex AI, Azure OpenAI
ML frameworks	TensorFlow, scikit-learn, XGBoost, LightGBM, ONNX
infrastructure	Kubernetes, Terraform, GitHub Actions, MLflow, Weights & Biases

github stats

contribution activity

my github contributions eating themselves

recent writeups

i write about what i'm building and learning. nothing polished, more like notes to my future self that happen to be public.

Multi Agent Ai Orchestration Patterns _2026-07-13	Model Context Protocol And Tool Use _2026-07-13
Clear Topic Name _2026-07-12	Multi Agent Llm Financial Trading Framework _2026-07-10