I am an AI/ML engineer specializing in LLM applications and production machine learning systems, currently building, maintaining, and shipping production LLM systems at FBS. I work primarily in Python and Java, with hands-on focus on large language model (LLM) applications, RAG-based systems, agentic systems, and machine learning.

I hold a MSc in Computer Science from NDSU (US) with focus on Machine Learning and a BSc in Computer Science from AIUB (BD).

• Languages: Python, Java
• AI/ML: Large language models (LLMs), [RAG / retrieval-augmented generation], machine learning, model deployment, [prompt engineering / fine-tuning]
• ML / LLM frameworks: PyTorch, TensorFlow, scikit-learn, Hugging Face Transformers, LangChain / LlamaIndex
• LLM tooling: Vector databases (Pinecone, Weaviate, FAISS, pgvector), OpenAI & Anthropic APIs, Ollama
• MLOps & infra: Docker, Kubernetes, CI/CD (GitHub Actions), MLflow, Weights & Biases

An open-source harness for converting PDFs to Markdown with any OpenAI-compatible vision model. Bring your own base_url, model, and key.

It adds page-level guardrails for repetition, truncation, text coverage, hallucinations, and refusals, with an escalation path that retries failures using reseeding, higher DPI, a fallback prompt, and finally the PDF text layer. Each page keeps its own confidence score, warnings, and source, so bad VLM output doesn’t silently make it into the Markdown.

Stack: Python · langchain-openai · pypdfium2 · Pydantic · markdown-it-py · Typer · asyncio · pytest · mypy · uv

A LangGraph ReAct agent with RAG over NIST CSF 2.0, NIST SP 800-53, and CIS Controls v8. Its tools run through a standalone MCP HTTP server and can also be called directly from MCP clients like Claude Desktop.

Includes Langfuse tracing, versioned prompts, RAGAS evals (0.97 faithfulness, 0.78 context recall), and GitHub Actions CI.

Stack: Python · LangChain · LangGraph · Groq · ChromaDB · HuggingFace TEI · MCP · Langfuse · RAGAS · Docker · uv