Headline

AI Architect | I take GenAI from impressive demo to production system people trust | Agentic AI · Multi-agent orchestration · GraphRAG · LLM evals · LangGraph · MCP · Azure

About

Every AI demo works. The real test is day 40, when actual users ask the questions nobody tested. That's where I come in. I'm an AI Architect at In Time Tec, and I design agentic and GraphRAG systems that still hold up after the demo ends. A few things I've learned the hard way: → The model is rarely the problem. Most "the AI got it wrong" bugs I've chased turned out to be retrieval, data or a missing tool, not the LLM. → Measure before you argue. If a design choice can't be settled with an eval, we're just debating opinions. → Autonomy is earned. I start agents on a short leash, measure how they behave, and widen their freedom as the evals prove they can handle it. → Agents should ask before they act. Anything that changes real data waits for a human to say yes. → Speed and cost are features. A great answer in 9 seconds loses to a good one in 2. I joined our AI practice on day one and helped build the team. I also sit on the sales side of the table, scoping and demoing live, which keeps my designs honest about budgets and deadlines. The part people don't expect: I wrote C# for six years before I touched an LLM. That's where I learned to care about what breaks at 2 a.m. Toolkit: LangGraph, MCP, Neo4j, Azure AI Foundry, Azure OpenAI, LlamaIndex, Langfuse, RAGAS, Python. Need an AI Architect who ships? hello@mustafamasvi.com

Location

Greater Bengaluru Area

AI stack

claude

hermes agent

langfuse

langgraph

ollama

openai

Work experience

Technical Architect (AI / GenAI) at In Time Tec, Dec 2022 - Present · 3 yrs 11 mos, Bangalore Urban, Karnataka, India, Chapter 3: Building an AI practice from the ground up. When In Time Tec started its AI practice, I was one of the first people in the room. Since then my job has shifted from "can we build this?" to "should we, and will it survive production?" What a typical week looks like: - A call with sales, turning a vague "we want AI" into a problem worth solving - Design reviews where I'm the one asking "how will we know it works?" - Sketching agent graphs on a whiteboard with engineers, then helping them get unstuck - Pushing for an eval before a feature, every single time The work spans agents, knowledge graphs, RAG and voice AI, for teams across Europe, the Middle East, Asia and Australia.

Senior Software Engineer at Odessa, Oct 2019 - Nov 2022 · 3 yrs 2 mos, Bangalore, Chapter 2: Learning that performance is a feature. I worked on a lease-management platform where a slow screen meant a frustrated finance team on the other end. I spent three years profiling, rewriting SQL and building code analysers that caught bugs before review did, and it earned me a move into the core platform team. This is where I stopped writing code that works and started writing code that lasts.

Software Engineer at In Time Tec, Jul 2016 - Oct 2019 · 3 yrs 4 mos, Bengaluru, Karnataka, India, Chapter 1: First job, first production bug. C# and ASP.NET, REST APIs for client integrations, and the humbling discovery that "works on my machine" is not a deployment plan.

Education

Bachelor of Engineering (BE), Information Technology, Dayananda Sagar Academy Of Technology And Management, Bangalore

R.V. PU College