An AI/ML Engineer - I take an idea, turn it into something real, and make it work in production.

I'm a software developer at the core. I've built in Python and C++ for years and committed fully to software/CS after Bachelor's in Electronics (2018), then from 2024 went deep on AI/ML: classical machine learning, deep learning, then LLMs, RAG, and agentic systems. I dig into how things work from first principles, then build production AI/ML systems end to end. I don't just prototype. I ship and show the work that works in the real world.

• ✔️ Agentic AI: LLM agents, multi-agent orchestration, MCP, tool-use
• ✔️ RAG and document-AI: private, local-first "chat with your data," grounded and citable
• ✔️ The ML and deep-learning foundations underneath: sound systems, not demos

I teach and mentor coders worldwide, from young first-timers to working adults, both as an instructor and on my channel. So I build code to be learned from, not just used:

🔹 Readable by design: clear READMEs and comments written to teach, so anyone can follow, run, and learn from the code 🔹 Modular, not one-off scripts: clean structure and pinned dependencies, easy to navigate and extend 🔹 Honest and reproducible: real results and trade-offs, not demos

Applied AI: agents, RAG, and document-AI

• folio-mcp: privacy-first, zero-cost "chat with your documents" assistant. RAG over local Ollama, MCP, CLI plus an OAuth web app, with an honest speed/quality benchmark.
• ollama-mcp-chat-cli: terminal MCP chat agent on local Ollama, with Groq/GitHub fallbacks.
• multi-provider-resume-extractor: typed multi-provider LLM client for schema-validated structured extraction from messy resume PDFs.
• ollama-litellm-tool-calling: fully local LLM tool-calling via litellm.

Machine learning and deep-learning foundations

• bert-imdb-sentiment: BERT fine-tuning for sentiment in PyTorch. Modular package plus CLI, principled early stopping, and honest metrics with caveats (F1 0.83 on held-out test, 0.86 on a truly-unseen slice).
• credit-risk-ml-pipeline: imbalance-aware credit-default risk. Leakage-safe pipeline, ROC AUC and PR-AUC instead of misleading accuracy, a tuned operating point, and decision-tree vs random-forest vs XGBoost compared honestly.
• covid-topic-modeling-faiss: LDA topic modeling over COVID-19 research papers with FAISS semantic search. Perplexity-based model selection, a real search CLI over the topic vectors, and the honest ceiling stated plainly.

Python · PyTorch · TensorFlow · scikit-learn · LangChain · LangGraph · MCP · RAG · Vector DBs · LiteLLM · LMStudio · Docker

Building production agentic AI ("governed autonomy") for financial services, deepening my ML and deep-learning foundations, and building in public.

• LinkedIn: https://www.linkedin.com/in/abd-ansari
• YouTube: https://www.youtube.com/@bitzntwist