Solution Architect | AI/ML & GenAI Engineer | Cloud Architect | Systems & Semiconductor R&D
I have 20+ years of experience designing scalable enterprise and cloud-native systems across AWS, Azure, and GCP. My current focus is on building AI-native platforms, custom LLM/Transformer architectures, agentic workflows, cloud modernization, and applied AI for enterprise domains such as retail, manufacturing, automotive, and document intelligence.
I like working at the boundary of software architecture, AI research, systems engineering, and semiconductor innovation.
- Custom Transformer and LLM architecture experiments focused on lower memory movement, smaller parameter count, and efficient inference.
- Agentic AI and multi-agent workflow design for software engineering, architecture generation, document processing, and enterprise automation.
- LLM-powered extraction, mapping, and structured JSON generation for enterprise document automation.
- Multi-agent solution patterns using planner agents, researcher agents, coding agents, reviewer agents, retrieval agents, critic/evaluator agents, and workflow orchestration.
- Advanced Reinforcement Learning concepts for autonomous decision-making, optimization, planning, reward modeling, and adaptive AI systems.
- Custom Transformer and LLM architecture experiments focused on lower memory movement, smaller parameter count, and efficient inference.
- LLM-powered extraction, mapping, and structured JSON generation for enterprise document automation.
- Exploring OS, semiconductor, and accelerator ideas for future AI systems.
- Applied ML using PyTorch, Scikit-learn, Hugging Face, Ollama, AWS Bedrock, Azure OpenAI, and Vertex AI.
- Creating cloud-native architectures that are secure, scalable, and cost-aware
- Multi-cloud solution architecture across AWS, Azure, and GCP.
- Serverless, ECS/Fargate, Lambda, API gateways, event-driven systems, DynamoDB, Firestore, BigQuery, and cloud-native integrations.
- Retail loyalty, campaign, voucher, customer savings, and large-scale transactional data systems.
- Exploring custom operating-system architecture, per-core scheduling, capability-based security, syscall gates, memory protection, and multi-architecture boot flows.
- Exploring custom ISA, AI accelerators, RISC-V/Shakti ecosystem, simulation, verification, and PDK-backed chip IP.
- Research interest in efficient compute architectures for LLMs and AI workloads.
- ML-assisted inverse design for silicon-based block copolymer self-assembly and lithography-oriented materials discovery.
| Area | Project | Description |
|---|---|---|
| LLM Research | PRISM-LLM | Memory-efficient LLM/Transformer experiments for low-VRAM training and inference. |
| OS Research | Bharat-OS | Experimental OS architecture with per-core scheduling, memory protection, capabilities, and multi-arch support. |
| Materials AI | nanobcp-ai | ML inverse-design engine for block copolymer self-assembly and semiconductor lithography research. |
| AI Resume Automation | Resume Formatter / AWS Bedrock Pipeline | Template manifest extraction, resume data mapping, LLM-based document formatting, and structured JSON generation. |
| AI Architect Studio | Nayvid AI Architect Studio | Multi-agent architecture workflow portal for requirement intake, solution design, and architecture brief generation. |
| Cloud Architecture | Retail Loyalty / Marketing Platforms | AWS/GCP/Azure enterprise systems for loyalty, customer data, serverless, and event-driven workloads. |
I am exploring alternatives and improvements to standard Transformer architectures with a focus on:
- Reducing parameter count without losing reasoning quality.
- Lowering data movement between memory and compute.
- Improving context handling using memory banks, gating, compression, and structured retrieval.
- Designing architectures that are practical for small GPUs and edge/low-resource environments.
- Studying mathematical alternatives to attention-heavy computation for LLMs, image generation, and video generation.
Research keywords:
Efficient Transformers · Memory-efficient Attention · GQA · KV Cache Optimization · Low-VRAM Training · AI Accelerators · Custom ISA · RISC-V
AI/ML & Agentic AI: PyTorch · Scikit-learn · Hugging Face · LangGraph · CrewAI · AutoGen · Ollama · AWS Bedrock · Azure OpenAI · Vertex AI
Cloud: AWS · Azure · GCP · Lambda · ECS/Fargate · Cloud Run · BigQuery · Firestore · DynamoDB
Languages: Python · Java · C# · JavaScript · TypeScript · C/C++
Architecture: Microservices · Event-driven Systems · Serverless · API Management · Distributed Systems
DevOps: Docker · Kubernetes · Jenkins · GitHub Actions · Bitbucket · Terraform/CloudFormation
R&D: Operating Systems · RISC-V · Custom ISA · AI Accelerators · VLSI Exploration
- Building AI-native engineering workflows.
- Improving resume/document automation using LLM-based manifests and deterministic rendering.
- Designing efficient custom LLM/Transformer architectures.
- Exploring OS, semiconductor, and accelerator ideas for future AI systems.
- Creating cloud-native architectures that are secure, scalable, and cost-aware.
| Area | Project | Description |
|---|---|---|
| LLM Research | PRISM-LLM | Memory-efficient LLM/Transformer experiments for low-VRAM training and inference. |
| OS Research | Bharat-OS | Experimental OS architecture with per-core scheduling, memory protection, capabilities, and multi-arch support. |
| Materials AI | nanobcp-ai | ML inverse-design engine for block copolymer self-assembly and semiconductor lithography research. |
| AI Resume Automation | Resume Formatter / AWS Bedrock Pipeline | Template manifest extraction, resume data mapping, LLM-based document formatting, and structured JSON generation. |
| AI Architect Studio | Nayvid AI Architect Studio | Multi-agent architecture workflow portal for requirement intake, solution design, and architecture brief generation. |
| Cloud Architecture | Retail Loyalty / Marketing Platforms | AWS/GCP/Azure enterprise systems for loyalty, customer data, serverless, and event-driven workloads. |
"Architecting cross-domain solutions where AI, cloud, digital transformation, business value, and next-generation compute come together."