🎓 B.Tech AI & ML Graduate | 🧠 AI & Web Developer | 🌐 Erode, India

I’m a B.Tech graduate in Artificial Intelligence and Machine Learning from Bannari Amman Institute of Technology.
Passionate about Machine Learning, NLP, Deep Learning, and Cloud Computing, I enjoy building AI-powered applications, automation tools, and cloud-integrated systems.
I’m proficient in C, Java, Python (Basics), SQL, and have hands-on experience with AI Agents, LLMs, Vision Models, and Cloud Technologies.

AI Engineering Intern — Wipro Connected Services (formerly Harman DTS) · Feb 2026 – Jun 2026

• Achieved 82% executable-SQL generation (37/45) on the Spider benchmark by building an NL2SQL2NL system on the Model Context Protocol (MCP) with schema-aware prompting, enabling context-aware, multi-turn plain-English querying of relational databases for business users.
• Reduced SQL composition time by 40% and passed 100% of an 8-query business demo suite by delivering the pipeline to non-technical stakeholders with conversational, Matplotlib-based chart generation.
• Deployed team-built agents and MCP servers to AWS Bedrock AgentCore Runtime by containerizing FastAPI services as linux/arm64 Docker images against the AgentCore service contract and publishing to Amazon ECR, covering both HTTP and MCP runtime protocols.
• Achieved 92% tool–agent alignment accuracy and 35% faster agent development by building a Multi-Agent Workflow Optimizer with ontology-driven requirement parsing and AgentBank validation, preventing 5+ duplicate tool implementations.

• 🔁 turnloop – Agentic Coding Harness (Python, Textual, vLLM, Modal, PyPI) — github.com/Pranesh-2005/turnloop
  Cut agent input-token usage by 20% and tool-call error rate to 0% by building an evaluation harness and running a 7-arm, 84-run ablation on agentic-harness design; disabling context compaction cost 39% more input tokens (273K vs 196K median). Shipped to PyPI — 21,300 lines of Python, 383 tests, only 5 runtime dependencies, no vendor SDKs — and self-hosted the 744B-parameter GLM-5.2 MoE evaluation target on vLLM across 4x H200 GPUs via Modal.
• ☁️ CloudLens – Multi-Agent AWS Operations Platform (LangGraph, FastAPI, A2A) — github.com/Pranesh-2005/Cloudlens
  Built a LangGraph supervisor routing to 6 specialist agents for cost, forecasting, security, and deployment across tenant AWS accounts, exposed through both an A2A protocol server and a FastMCP endpoint, with Fernet-encrypted credentials and human approval gating on every mutating action.
• 🔍 RagObserve – RAG Observability Platform (Python, FastAPI, SQLite, PyPI) — github.com/Pranesh-2005/RagObserve
  Cut tracing overhead 136x (379.8 ms → 2.8 ms per trace) by moving the SQLite writer to WAL journaling, making the tracer safe on an application's hot path; covers 11 LLM providers with LangChain and LlamaIndex adapters, fully offline.
• 🧠 LLM Chatbot with Long-Term RAG Memory (Qwen2.5-7B, LoRA, Unsloth, Modal, GGUF)
  Fine-tuned Qwen2.5-7B-Instruct with LoRA (Unsloth) on serverless Modal A10G GPUs using 14K instruction-tuning samples engineered from 120K+ raw chat messages; added long-term memory via RAG retrieval over 30K+ embedded bursts and exported 4-bit GGUF models for local inference.

• Anthropic: Introduction to Model Context Protocol; Model Context Protocol: Advanced Topics
• Microsoft: Azure AI Fundamentals (AI-901)
• freeCodeCamp: Machine Learning with Python

• ⭐ github-readme-stats-fast — GitHub profile stats generator with a companion GitHub Action: 270+ stars, 59 forks.
• 📦 PyPI packages — turnloop (agentic coding CLI), guardix (prompt-injection guard), Text-Emotion, DocTok.
• 🤗 35+ AI demos built and deployed on Hugging Face, spanning RAG, agents, and model-internals visualizers (GPT-2, ViT, YOLO).

• Artificial Intelligence & Machine Learning
• Generative AI & LLMs
• AI Agents & Automation
• Cloud Computing (Azure) & Docker
• Database Design & Vector Databases (ChromaDB)
• Intelligent Automation

💻 Always open to collaborating on exciting **AI, ML, or Cloud-based projects**. Let’s innovate together!