About me
Hi, I’m Gaurav Pooniwala — a founder and AI engineer focused on making LLM-powered systems reliable in high-stakes workflows.
Over the past decade, I’ve worked across research labs, startups, and hypergrowth environments: from deep learning R&D at Samsung to building GenAI video systems used by tens of millions of people, and leading AI initiatives in fintech and logistics.
Today, my work centres on LLM observability, human-in-the-loop (HITL) review, and how teams can design workflows that balance automation with judgment — without burning out reviewers or putting users at risk.
Current focus
I’m currently building and exploring tools around LLM reliability and observability: analysing failure modes, designing evaluation strategies, and automating parts of the human review loop for teams who operate in regulated or high-impact domains.
Alongside that, I partner with teams as a consultant, advisor, and senior IC on projects where LLM workflows need to move from promising prototype to dependable production systems.
I’m open to selective opportunities — including senior IC, leadership, consulting, advisory work and speaking roles — where I can help teams design and ship reliable AI workflows end to end.
How I work
- Start from the problem and workflow. I begin by understanding the real process, who is involved, and what a good outcome looks like, before choosing tools or architectures.
- Align early with stakeholders. I work closely with product, operations, and leadership to agree on goals, constraints, and trade offs so the solution fits the team and the business.
- Ship small, learn fast. I prefer minimal versions that can be tested in reality, then iterate based on user feedback, data, and what actually breaks.
- Design systems that can be owned and debugged. I care about clarity of behavior, observability, and simple mental models so that teams can understand, maintain, and improve what we build.
Capabilities
- LLM and agentic system design I design assistants, tools, and multi step workflows that use LLMs and other models to solve concrete problems for users and teams.
- Workflow and automation design I map current processes, decide what should be automated and what should stay human, and shape end to end flows that feel natural to use.
- Evaluation, safety and reliability I define what success looks like, set up checks and feedback loops, and think about failure modes so that systems behave in a predictable way.
- Strategy and execution I bridge between engineering, product, and leadership so that ideas turn into shipped products, with clear responsibilities and sensible trade offs.
In past roles I have used tools like Python, PyTorch, LangChain, Hugging Face, and common cloud and MLOps stacks, but I treat these as implementation details rather than the main story.