I build production-oriented AI systems — from LLM fine-tuning pipelines to scalable multi-agent platforms.
Strong focus on real-world deployment, backend orchestration, and explainable AI products.

🌐 Portfolio: sadmansakibrafi.vercel.app
📩 Email: sadmansakibrafi.hey@gmail.com

• Large Language Models (LLMs): fine-tuning, prompting, evaluation
• Multi-agent AI systems & orchestration
• AI-driven backend platforms (FastAPI, NestJS)
• Vector search & semantic retrieval (Pinecone, FAISS, Qdrant)
• Production-ready AI pipelines (local + cloud)
• Explainable & scalable AI products

• Supervised Learning, Classification, NLP
• Large Language Models (LLMs)
• Prompt Engineering & SFT
• LoRA / Parameter-Efficient Fine-Tuning

Tools & Libraries:
Hugging Face, Scikit-learn, LangChain, CrewAI, Ollama, Gemini

• FastAPI, NestJS, ASP.NET
• REST API design & integration
• Context memory, tool calling, agent routing
• Vector embeddings & similarity search

• Python (primary)
• JavaScript / TypeScript
• C++, C#

Backend-focused AI assistant platform built with FastAPI.
Implements agent routing, short-term memory, vector embeddings, and LLM abstraction layers for scalable reasoning.

AI-driven recruitment system that semantically analyzes and ranks resumes against job requirements.
Supports batch CV ingestion, gap detection, and explainable scoring using a hybrid LLM pipeline (Gemini + Ollama) and Pinecone.

API-based AI service that generates contextual social media captions for Facebook, Instagram, and Twitter.

Multi-agent tutorial generation system using CrewAI and Ollama-Mistral.
Designed for local inference, privacy, and fast iteration.

Ongoing project: real-time AI assistant for automated business interactions.
Focuses on backend orchestration, context memory, tool calling, and extensible APIs for chat and voice.

• Supervised Fine-Tuning (LLMs):
  Fine-tuned Qwen3-1.7B-Instruct to convert unstructured sales data into structured lead intelligence.

• LoRA Fine-Tuning (NLP):
  Fine-tuned DistilRoBERTa on IMDB using LoRA for efficient sentiment classification.

BSc in Computer Science & Engineering (Software Engineering)
American International University-Bangladesh (AIUB)
GPA: 3.64 / 4.00