I design and ship LLM-powered systems that get embedded straight into real workflows — turning messy business problems into production-grade AI, one retrieval pipeline and autonomous agent at a time.

📍 Pune, India  |  3+ years shipping GenAI systems at Amdocs

I sit at the intersection of applied ML and client delivery — the "forward deployed" part means I go where the problem is, not just where the model is. That's meant:

• 🔁 Architecting RAG pipelines that ground LLMs in private, domain-specific data — without leaking that data to external services
• 🤖 Building multi-agent systems (LangChain Agents, AutoGen, CrewAI) that plan, reason, and call tools autonomously
• ⚙️ Shipping the boring-but-critical infra: Flask/FastAPI microservices, caching, parallel inference, monitoring for hallucinations
• 🗣️ Translating what a stakeholder says they want into what the system should actually do

Impact I can point to: 50% reduction in manual effort across target workflows · ~35% lower inference latency · 40–45% lift in factual response quality through retrieval + prompt optimization.

Solr AI Query & Update Studio — a natural-language layer over Apache Solr. Type a plain-English request, get an optimized Solr query and synchronized updates propagated across Unix XML files and Oracle databases — no hand-written query syntax required. Runs on a local LLM via Ollama, ships with a browser extension for in-context use, and auto-generates HTML audit reports for every change.

Python Ollama Apache Solr Oracle Java Serialization Chrome/Firefox Extension APIs

GenAI / Agentic AI

Vector Search & Embeddings

Languages & Backend

Data & DevOps

• Deepening advanced RAG techniques — hybrid search, re-ranking, query rewriting
• Building scalable multi-agent orchestration for enterprise use cases with AutoGen & CrewAI
• Contributing to open-source LangChain extensions and RAG optimization repos

💬 Open to conversations on GenAI systems, agentic architectures, or forward-deployed engineering — reach out anytime.