"The Analytical Engine weaves algebraic patterns, just as the Jacquard loom weaves flowers and leaves." — Ada Lovelace, 1843 · the first person to see that machines could manipulate meaning, not just numbers. I build the descendants of that idea.
class Sarvesh(GenAIEngineer):
role = "AI Full-Stack Engineer @ Aditya Birla Group"
experience = "4+ years building backends → enterprise platforms → production AI"
obsession = "making LLMs reason reliably, cite their sources, and never hallucinate"
off_screen = ["philosophy", "history of ideas", "where the two meet code"]
def believe(self) -> str:
return "Code is applied philosophy — every abstraction is a stance on what matters."I'm a GenAI engineer who treats an agent's control flow the way a philosopher treats an argument: every branch is a claim, every tool-call a piece of evidence, every guardrail a refusal to fool yourself. I architect multi-tenant RAG systems, stateful agentic pipelines, and the LLMOps plumbing that keeps them honest in production.
|
🧠 Agentic AI & LLMs Multi-agent coordination · tool-calling loops · prompt engineering · system-prompt guardrails · human-in-the-loop |
🔁 RAG, Retrieval & Vision Semantic KB design · contextual re-ranking · adaptive chunking · multimodal layout analysis · SSE streaming |
- 🤖 Enterprise multi-agent platforms — LangGraph state machines with conditional query routing, semantic retrieval, and source-grounded loops that drive hallucinations toward zero.
- 📚 Multi-format ingestion engines — 8 specialized parsers + an adaptive chunker, offloading heavy parsing to distributed Celery Chords so nothing times out.
- 🔌 MCP tool orchestration — letting LLMs safely read, search, and map external ecosystems (Drive, Notion) through the Model Context Protocol.
- 🔭 LLMOps observability — LangSmith trace-linking, token/cost auditing, and high-throughput FastAPI SSE streaming for live agent traffic.
"No one ever steps in the same river twice — for it is not the same river, and they are not the same person." — Heraclitus. The best mental model I've found for an agent's state across a long conversation: every turn, the river and the swimmer have changed.
"We can only see a short distance ahead, but we can see plenty there that needs to be done." — Alan Turing, 1950. Still the truest sentence ever written about working in AI.
Between commits I read about the history of ideas — from the library of Alexandria to the Macy conferences that birthed cybernetics — because every "new" frontier in AI is usually an old human question wearing better math.
