A grounded question-answering system for Armada’s internal Customer Support team, searching runbooks, help-center content and vendor documentation.
Sole owner, from discussions with the support team and an empty repository through production integration.
An on-device RAG application designed for document question answering on Armada edge hardware without internet access.
Sole engineering owner across document ingestion, local model serving, retrieval and cited question answering.
Engine-independent predictions of KV-cache capacity, latency, throughput and serving cost from spec sheets, with optional GPU-profile calibration.
Built the planner and used its predictions to choose serving configurations, then deployed them. Spec-sheet KV-cache capacity predictions were exact on 25/25 dense and MoE configurations.
With brief GPU profiling, dense mean error falls from 15.9% to 3.6% for TTFT (n=174) and 13.8% to 5.2% for batched TPOT (n=242). Errors are mean absolute percentages; profiled results are partly in-sample.
Contributions to an inherited RAG application, distinct from the Support Agent I later owned.
Contributor. Redesigned the LangGraph workflow and improved context preparation, retrieval and metadata handling.
A lifestyle-image generation pipeline that preserves the product while adapting its appearance to a new scene.
Built the Flux and ControlNet pipeline; modified diffusion sampling and product-relighting work were both adopted into Avataar’s production pipeline.
Sparse-image-conditioned novel-view synthesis for 3D reconstruction at Avataar AI.
Fine-tuned sparse-image-conditioned diffusion models and worked on conditioning, distributed training and SLURM, building on a senior engineer’s initial setup.
Video classification for cricket shots, from frame extraction to a runnable prediction demo.
Built embedding-based video classifiers and fine-tuned VideoMAE.
A pipeline for comparing and substituting document-processing and retrieval models to select suitable components.
Evaluated table-structure models, OCR models, layout detectors and embedding models, and built a pipeline that supports switching between candidates.
Virtual try-on using Florence2 and IDM-VTON, with a three-stage CAT-TryOff pipeline and a single-stage Any2Any-Tryon/FLUX approach.