Software Engineer & Researcher focused on building intelligent systems and practical AI for real-world problems. Currently pursuing my Master's at the University of Utah, working on generative and interpretable AI for biomedical signals.

Portfolio: deekshith.me — highlighted work and writing, including the KV-cache blog.

• mlx_sam3 — SAM 3 port to Apple MLX with lightweight prompting and fast inference experiments.
• mlx_perception_models — perception model experiments optimized for Apple MLX workflows.
• Synthetic Road Sign Detection — synthetic data generation in Blender with DETR/YOLOv5 training and measurable IoU gains.
• Mars Captioning UI — CLIP-based captioning and semantic search over Mars rover imagery.

• Augment (Code) — agentic mind-map engine that captures thoughts, embeds context, and runs multi-tool workflows with memory.
• Spot Reader — AI-enhanced EPUB reader with RAG, chat, annotations, and streaming responses (Next.js + FastAPI + Supabase + AWS).
• LangGraph Story Agent — multi-agent creative writing system with persona-driven story generation.

• ECG Deep Learning @ SCI Institute — contrastive pretraining on 1M+ ECGs, distributed training pipelines, and generative interpretability work.
• Long Document Transformers — reproducible benchmarks for long-form document classification using PyTorch DDP + HuggingFace.

• KV-Cache Blog — practical deep dive on KV-cache optimization and inference tricks for LLMs.
• MLX SAM3 Worklog — notes on porting SAM3 to MLX on Apple Silicon with a demo app.

Languages: Python C++ Java TypeScript Go

Frontend: React Next.js Tailwind shadcn Jest

Backend: FastAPI Flask Spring Boot Kafka Node.js PostgreSQL Redis SQLAlchemy Docker

AI/ML: PyTorch Transformers LangGraph Diffusion GANs Contrastive Learning CLIP CrewAI

Infra & DevOps: AWS EC2 S3 Nginx Docker Compose CI/CD GitHub Actions Supabase

Databases & Graphs: Neo4j Pinecone Firebase Prisma

• On-device multimodal models for Apple Silicon
• Building AI that organizes thought and context at scale
• Signal-level generative modeling for interpretable biomedical AI
• Multi-agent orchestration with LangGraph
• Graph-RAG systems with structured memory

• 🌐 deekshith-dade.github.io
• 🧳 LinkedIn
• ✉️ Email