I fine-tune models, wire up RAG pipelines, build multi-agent systems, and deploy them.
I build AI systems end-to-end — QLoRA fine-tuning, RAG pipelines, multi-agent workflows, computer vision, MLOps. I care about the whole picture: data, modeling, APIs, experiment tracking, cloud. Finishing my B.E. in CS (AI & ML) and looking for a role where I can keep building.
# mohammed_omer.py
class MohammedAbdulOmer:
def __init__(self):
self.name = "Mohammed Abdul Omer"
self.location = "Hyderabad, India 🇮🇳"
self.education = "B.E. Computer Science (AI & ML) — Lords Institute of Engineering"
self.focus = ["LLM Engineering", "RAG Pipelines", "LLM Fine-tuning", "MLOps", "Computer Vision"]
self.stack = ["Python", "PyTorch", "LangChain", "LangGraph", "CrewAI", "FastAPI", "HuggingFace"]
self.status = "Open to AI / ML Engineer roles 🚀"
def say_hi(self):
print("Let's build something intelligent together!")
me = MohammedAbdulOmer()
me.say_hi()| Area | What I Build |
|---|---|
| 🎯 AI Routing & Orchestration | Multi-provider AI routers, task classification, benchmark pipelines |
| 🔍 Advanced RAG | Hybrid retrieval (BM25 + vector + re-ranking), agentic pipelines, multi-document QA |
| 🤖 LLM Engineering | QLoRA/PEFT fine-tuning, prompt orchestration, Groq / Gemini / OpenRouter APIs |
| 🧠 Agentic AI | Multi-agent systems, CrewAI, LangGraph ReAct agents, task planning |
| ⚙️ MLOps | MLflow experiment tracking, Evidently AI drift monitoring, W&B, reproducible pipelines |
| 👁️ Computer Vision | Object detection (YOLOv8), medical imaging, Grad-CAM explainability |
| ⚡ API & Deployment | FastAPI, Django, Docker, Railway, Render, HuggingFace Spaces |
|
Routes tasks to the right model — Groq, Gemini, OpenRouter, or Ollama — based on whether you care more about speed, quality, cost, or keeping data local. Classifies across 10 task types, supports 17+ models, has a benchmark mode to verify routing decisions, and falls back automatically when a provider fails. Full CLI included.
|
Three agents — researcher, analyst, writer — split up a research task, pull live web data via Tavily, and hand off work through CrewAI until a structured report comes out. You give it a topic; you get a report.
|
|
Fine-tuned Mistral-7B-Instruct-v0.2 on a medical Q&A dataset using QLoRA/PEFT on a Kaggle T4 GPU. Most of the actual work was sorting out quantization compatibility across library versions. Model and Gradio demo are live on HuggingFace.
|
RAG chatbot built around a LangGraph ReAct agent with hybrid retrieval — ChromaDB for vector search, BM25 for keyword matching, FlashRank for re-ranking. FastAPI handles streaming over SSE; Groq's llama-3.1-8b-instant does the generation. Streamlit and Gradio frontends included.
🔗 GitHub |
More projects on my →
Want to talk AI systems or discuss roles? →
