Building intelligent AI systems for real-world impact — from RAG pipelines to autonomous agents.

I'm an AI Engineer focused on turning Generative AI and LLM research into production-ready systems — retrieval-augmented generation, multi-mode AI agents, and backend infrastructure that supports them. I'm a 2026 BE graduate in Electronics & Communication, and I've worked as an AI & Web Development Intern at Unidemy Global, where I built and shipped applied AI tools rather than just prototypes.

I care about systems that are useful in production, not just impressive in a demo — clean APIs, sensible data pipelines, and agent behavior that's predictable enough to trust.

🔎 AI-Web-Agent — Multi-Mode AI Research Assistant A multi-mode research assistant (Fast / Smart Search / Deep Research) on a React + FastAPI stack, with a query-classification layer that routes by complexity — cutting unnecessary compute on simple queries by ~40%. Runs local LLM inference via Ollama (Qwen 2.5 7B) with live web search tool-calling for source-aware, evidence-backed answers, plus hallucination-reduction mechanisms (future-event detection, unsupported-claim rejection) cutting fabricated responses by ~45% vs. baseline recall. Fully containerized (Docker Compose) and deployed on Lightning AI with GPU acceleration. Python FastAPI React Ollama Docker

🤖 Composio_Research_agent — Automated SaaS API Discovery A production-grade research pipeline (clean domain/use-case/adapter architecture) automating discovery of API capabilities, auth models, and MCP support across 100+ applications — cutting manual research time by ~70%. Includes an independent verification agent that re-fetches cited docs to score each field Pass/Fail/Unknown (defaulting to Unknown on weak evidence, to eliminate fabricated metadata), plus a live analytics dashboard with 8 charts. Python OpenAI API Composio Jinja2 Chart.js

📚 rag-customer-support-chatbot — Enterprise Knowledge Retrieval A RAG chatbot grounding answers in company policy documents using FAISS vector search and Hugging Face sentence embeddings, feeding top-k retrieved chunks into a locally hosted Ollama (Gemma 3:1B) model — reducing hallucinated answers by an estimated ~50% vs. a no-retrieval baseline. Delivered via a conversational Streamlit UI with multi-turn memory across 4+ document categories. Python LangChain FAISS Streamlit

📊 Data & Analytics Foundations Beyond AI systems, I also work end-to-end with data: electronics-products-eda-mysql (SQL-based EDA & feature engineering) and powerbi-customer-churn-analysis (churn analysis dashboarding) — the data layer that feeds into good AI systems.

Languages

AI / ML & Generative AI LLMs (Qwen · Gemma · LLaMA) · Vector Databases · Sentence Transformers · Prompt Engineering · NLP · Tool Calling · Hallucination Reduction · Stable Diffusion · DreamBooth · ControlNet

Backend & APIs REST APIs

Cloud & DevOps

Data & Analytics

• Multi-agent orchestration frameworks (LangGraph / agent-to-agent workflows)
• Production-grade RAG evaluation & retrieval quality
• Scalable backend architecture for LLM-serving APIs

Open to AI Engineer / GenAI / LLM Systems roles — Bengaluru (On-site · Hybrid · Remote). Reach me on LinkedIn or via email above.