I work with LLMs and production ML systems. Mostly tinkering with RAG, fine-tuning models, and trying to make AI actually work in the real world instead of just in Jupyter notebooks.
Currently at building GenAI stuff for financial data.
LLMs: LangChain, fine-tuned LLaMA, GPT-4, Claude, Gemini
ML/Data: PyTorch, TensorFlow, transformers, scikit-learn
MLOps: ZenML, DVC, MLflow - trying to make ML reproducible
Cloud: Azure, GCP
Languages: Python mostly, some SQL and bash
Finetuning-Llama-3.1-Unsloth
Fine-tuned LLaMA 3.1 using Unsloth. It's basically 60% faster than standard PEFT, which is pretty cool. Useful if you're doing domain-specific LLM stuff.
PDF-ChatBot-Langchain
RAG system for PDF Q&A. Proper source attribution (important for actual use). Has conversation memory too.
Chainlit-langchain-app
LLM chatbot interface. Built this to understand how to go from prototype to something you can actually deploy.
Mlops-project-zenml
ML pipeline orchestration with ZenML. Focused on making the whole thing reproducible and automated.
Mlops-DVC-StackOverflow
Data versioning + experiment tracking + CI/CD pipeline. Kind of a complete MLOps workflow.
BERT-Pretraining-finetuning
BERT from scratch with MLM pre-training, then fine-tuned on sentiment. Learned a lot about how transformers actually work by doing this.
Design-Patterns
Implemented Gang of Four patterns in Python. Useful to have this reference around.
Worked on financial data extraction with LLMs - took manual processes from weeks down to minutes. Built RAG systems that actually work in production. Spent time optimizing inference, dealing with vector databases, all that jazz.
Before all this, I was in financial operations (State Street) doing portfolio reconciliation, which is where I got interested in automation and data systems.
- Better RAG patterns (retrieval is still messy)
- Agentic workflows (LangGraph is interesting)
- LLM optimization (quantization, distillation, caching)
- Evaluating GenAI systems properly
📧 abhilashk.isme1517@gmail.com
🔗 linkedin
🌐 portfolio
Open to talking about GenAI, MLOps, building production systems. Always up for discussing weird ML problems too.