AI/LLM Technical Lead · Forward Deployed Engineer · Production AI Systems at Scale
📍 Manchester, UK · 🏢 Microsoft Industry Solutions Delivery
I build and ship production-grade AI systems for enterprise customers across financial services, sports, construction, and energy industries. My work sits at the intersection of customer problem solving and frontier AI deployment — from discovery and architecture through to production and evaluation.
At Microsoft's Industry Solutions Delivery team, I embed directly with enterprise customers to co-design and build AI applications.
- Multi-agent systems — parallel agent architectures, tool routing, Semantic Kernel, Microsoft Agent Framework
- RAG pipelines — hybrid search, semantic reranking, query planning, document intelligence
- MCP (Model Context Protocol) — building and integrating MCP tools and servers for real-time agent data retrieval
- Evaluation frameworks — LLM-as-judge, golden datasets, CI/CD integrated quality monitoring, Azure AI Evaluation SDK
- Production LLM deployment — prompt engineering, grounding, PII detection, content safety, session management
- LLMOps — observability, autoscaling, Redis caching, performance tuning at scale
| Area | Tools |
|---|---|
| Daily driver | Visual Studio Code , Claude models — agent mode, MCP, skills |
| LLMs | GPT models, Claude models |
| Orchestration | Semantic Kernel, Microsoft Agent Framework, LangChain |
| Search & Retrieval | Azure AI Search (hybrid + semantic), Vector DBs, Redis |
| Evaluation | Azure AI Evaluation SDK, LLM-as-judge, Prompt Flow |
Languages:
Python · C#/.NET · TypeScript · JavaScript · SQL
AI & Cloud:
Azure OpenAI · Azure AI Search · Document Intelligence · AI Content Safety
Container Apps · Azure Functions · APIM · App Gateway (WAF) · Cosmos DB · Redis
DevOps & Tools:
Azure DevOps · GitHub Actions · Docker · Application Insights
Sports Platform — Multi-Agent LLM System
Agentic RAG chat system handling thousands of RPS across catalogue and conversational queries. Parallel agent architecture with MCP tools for real-time player and club data retrieval. Custom evaluation framework baked into CI/CD pipeline.
Construction — AI-Powered RFP Bidding System
Intelligent document processing with geospatial RAG, winning proposal analysis, and conversational interface. Background evaluation pipeline via queue-based processing for continuous quality monitoring.
Financial Services — Enterprise RAG Chatbot
End-to-end RAG implementation with Azure AI Search and Semantic Kernel. Defined evaluation golden datasets with stakeholders and built LLM-as-judge scoring for safe enterprise deployment.
Energy — AI Timesheet Validation Workflow
AI-driven automation for procurement timesheet validation at one of the world's largest oil corporations.
- I've taken customers from "is this even possible?" to production — across PoCs, greenfield builds, and legacy integrations
- I build evaluation frameworks first — not as an afterthought
- I push back on customers when needed and define what "good" actually looks like
- I work at both ends: whiteboard and terminal
- MCP server design patterns and tool boundary architecture in agentic systems
- Evaluation strategies for text-to-SQL and open-ended LLM responses
- AI safety and interpretability in enterprise deployment contexts
- ✉️ vijaycse@live.in
⚡ Fun fact: The hardest part of enterprise AI isn't the model — it's getting the right people in the room to define what good looks like.