Who I am
My name is Tharun Chowdary Malepati. I work in AI and machine learning engineering, with a growing focus on generative AI and agentic systems. Engineerious is the public desk where I can organize what I’m building, what I’m ready to explain, and the questions I want to keep working on.
Engineer first. Curator second. Creator always.
What I work on
My interests span the full path from a model call to a system someone can operate and evaluate. These are the areas currently on the desk:
- LLM systems
- LLM applications · RAG · AI agents · multi-agent systems · MCP · A2A
- Retrieval and quality
- embeddings · retrieval · reranking · LLM evaluation · AI observability
- Adaptation and inference
- fine-tuning · LoRA and QLoRA · PEFT · quantization · vLLM · GPU inference
- Application engineering
- Python · FastAPI · LangChain · LangGraph · AutoGen · vector databases
What I’m trying to understand
I care about the gap between a clean AI demo and a useful system. How should retrieval be designed? When does an agent architecture earn its complexity? How do we evaluate behavior instead of admiring one good run? What changes when latency, cost, observability, and inference enter the picture?
Those questions matter more to me than repeating every announcement. They also give this site its shape.
Why I write
Learning becomes stronger when I can explain it simply. I want a useful entry to go beyond what happened and answer why it matters, how it works, who should care, and what an engineer should take away from it.
That means leaving room for uncertainty, mistakes, tradeoffs, and opinions. It also means being clear about what I tested, what I only researched, and what still needs work.
What you’ll find here
- Reviewed writing about AI engineering and system design.
- Daily intelligence and curated developments as they are approved.
- Project case studies centered on architecture and decisions.
- Practical notes on concepts, models, and frameworks when the material is ready.
Publishing standard
- AI-generated work stays unpublished until a named human reviews it.
- Reported, tested, and opinionated claims remain distinct.
- Primary sources and reproducible details are preferred.
- Unconfirmed projects and personal experience do not become placeholder copy.
- Corrections are part of the record.