aritro shome

AI4Bharat, IIT Madras

• investigating mechanistic interpretability and circuit tracing in SoTA open models (Gemma, Qwen) on reasoning tasks across modalities, languages and model scales, focused on vision-language models over Indic inputs.
• ran 40,000+ instrumented forward passes for activation capture on multi-billion-parameter models under a hard 24 GB VRAM ceiling, with component-attribution scripts to isolate critical circuits before committing to full runs.
• engineered a memory-efficient activation-patching framework: disabled fused and attention-kernel optimisations to expose per-layer activations, then managed the non-linear VRAM growth that followed.

stackPythonPyTorchHugging FaceCUDAmixed precisionLinux

Sarvam AI, media & dubbing

• built an automated evaluation framework for an NVIDIA lip-sync model; the pipeline caught and resolved a bug causing ~500% variance in output video file sizes.
• redesigned the production dubbing pipeline, replacing brittle hard-coded prompt harnesses with modular YAML instruction blocks that let the model's chain-of-thought infer context. translation quality went up; i picked the final configuration by a/b test.

stackPythonYAMLdata pipelinesLLM APIsa/b testingGit

seepipeline in production (video)

• hand-designed an interpretable 2-layer transformer that does two-digit addition with manually-set weights and no gradient descent, building attention and MLP computation from first principles to show exactly what each component contributes.

stackPythonPyTorchNumPy

seewriteupinteractive demo

• replicated Anthropic's persona / emotion-vector methodology to extract a steering vector for each of the seven deadly sins in Gemma, on a single consumer GPU.
• verified vector correctness through targeted steering and activation patching rather than eyeballing generations.

stackPythonPyTorchHugging FaceGemma

seewriteup

• built a modular Python engine that generates alpona, a Bengali folk art, through fractal-based layering, with an extensible OOP style-registration system.
• published a 13,000-image synthetic dataset at 1024×1024 on Kaggle for generative-ai research.

stackPythonNumPyOOP design

seegithubdataset