For the past year, I've been deep in the trenches of LLM post-training, reinforcement learning, and evaluation pipelines. I've built tools, run benchmarks, and fine-tuned models using the latest open-source stacks.
I like exploring whatever feels interesting at the moment—papers, systems, and implementations—and turning it into something concrete.
LiminalLearning is my public lab notebook for that: experiments, notes, and small builds I’d actually want to use.
The Bench
Rigorous logs of my experiments with various Post Training Methods (SFT, DPO, PPO, GRPO, etc) and various papers of interest. Documenting the messy reality of getting loss curves to converge.
First Principles
Deconstructing papers and building small implementations from scratch when it helps me understand.
The Timeline
Joined SAP R&I Lab
Started internship focusing on Fine-tuning LLMs for targeted Enterprise benefit.
Deep Dive: Post-Training
Spent the year building Post Training workflows and evaluation tooling.
Final Year @ UCLA
Finishing a BS in Computer Science + academic research in Model Performance
The Liminal Crossing
Exploring my interests from First-principles - and sharing what I learnt along the way