I’m an AI Engineer working across computer vision, multimodal learning, generative AI, and application engineering, with a focus on evaluation, deployment, and practical inference constraints.
About
I work on AI systems where model quality is only one part of the engineering problem. My focus spans evaluation pipelines, multimodal modeling, inference, application integration, and deployment.
Recent work includes cross-dataset deepfake detection, browser-native speech transcription and translation, and interpretable image classification with portable offline inference.
Projects
Built a multimodal deepfake video detector that combines visual representations with rPPG-derived physiological signals across multiple temporal windows. The system uses SigLIP2, CNN-Transformer encoding, and quality-aware cross-attention, with leakage-aware training and zero-shot evaluation across CelebDF v2, FaceForensics++ C23, and WildDeepfake.
95.73%In-Domain Accuracy
79.10%Cross-Dataset Macro
3Datasets Evaluated
Built a browser-only speech transcription and English translation application using Whisper, Transformers.js, ONNX Runtime Web, and Silero VAD. The inference path adapts between WebGPU and WASM, runs through WebWorkers, bounds memory use for longer audio, and supports timestamped editing, microphone and file input, and offline caching for privacy-preserving client-side execution.
BrowserClient-Side Inference
WebGPU/WASMAdaptive Runtime
OfflineCached Execution
Built an end-to-end binary classifier for benign versus malignant oral lesions using a fine-tuned DINOv2 and EfficientNetV2-B3 ensemble. Added Grad-CAM explanations for qualitative review, exported the ensemble to ONNX, and packaged the inference pipeline as an offline PyQt6 desktop application for portable use on local hardware.
0.96Test AUC-ROC
ONNXPortable Inference
OfflineRuns On-Device
Skills
Python, TypeScript, SQL.
LLM APIs, retrieval-augmented generation, embeddings, retrieval and reranking.
Structured outputs, tool calling, AI agents, LLM evaluation and observability, multimodal AI.
PyTorch, YOLO, OpenCV.
React, Next.js, FastAPI, PostgreSQL, pgvector, Redis, Celery.
Docker, Vercel, AWS, GitHub Actions.