• 🎓 Pursuing B.Tech in Electronics & Communication Engineering at LNMIIT (2023–27)
• 🤖 Agentic AI & LLMOps Engineer specializing in multi-agent systems and RAG pipelines
• 🔬 Exploring LLMs, Deep Learning, Embedded Systems, and Full-Stack AI Development
• 💡 Passionate about building production-grade AI systems, voice interfaces, and agentic frameworks
• 🌱 Currently diving deeper into evaluation frameworks, VLMs, and MLOps
• ✍️ Documenting my journey and projects here on GitHub

Ishan Bansal, Vidit Vinarma, Kartik Sharma — Advisor: Dr. Sandeep Saini, LNMIIT

• AI/ML Tech: BiLSTM, MediaPipe Pose Landmarks, Edge Deployment
• Architected a BiLSTM classifier (~110K parameters) achieving 94.89% accuracy and 0.945 macro F1 across 7 clinical classes, optimized for low-power CPU edge deployment
• Geometric post-processing engine sustaining 96.64% fall detection recall with 2.1% FPR and <2ms CPU latency at 15 FPS

AI-powered autonomous pipeline that produces full blog posts using a 5-agent LangGraph architecture with concurrent section writing.

🎯 What it does:

• Orchestrates 5 specialized agents for research, outlining, writing, editing, and publishing
• Delivers 5× faster content production vs. sequential generation using LangGraph's Send() fan-out pattern
• Zero schema validation errors in production via strict Pydantic v2 structured outputs

🛠️ Tech Stack:

• Orchestration: LangGraph, LangChain
• AI Models: Mistral AI, Google Gemini, Tavily Search
• Backend: Python, Pydantic v2, Streamlit
• Architecture: Multi-agent fan-out with exponential-backoff error handling

✨ Key Features:

• 🧠 5-agent autonomous pipeline with parallel concurrent writing
• 🔍 3-mode research system powered by Tavily web search
• 🛡️ Zero crash guarantee with exponential-backoff on API failures
• 📋 Strict structured outputs via Pydantic v2 across all agent nodes

AI-powered agricultural assistant with market prices, weather insights, and Hindi-first multilingual voice I/O.

🎯 What it does:

• Connects farmers to live market prices, weather forecasting, and go/no-go planting decisions via voice
• Supports Hindi and English speech with real-time ASR and TTS
• Persists agronomy knowledge into a local vector database for low-latency RAG

🛠️ Tech Stack:

• Backend: FastAPI, Python, APScheduler
• AI/ML: Mistral AI, OpenAI Whisper (ASR), Microsoft Edge-TTS, spaCy NER
• Vector DB: ChromaDB with all-MiniLM-L6-v2 embeddings
• APIs: Agmarknet market endpoints, Open-Meteo weather API

✨ Key AI/ML Features:

• 🗣️ Hindi-first bilingual voice interface with Whisper ASR + Edge-TTS
• 🌦️ Live weather forecasting with go/no-go planting engine
• 📦 Overlapping chunking + RAG pipeline for agronomy knowledge base
• ⏰ Automated hourly alert jobs via APScheduler background scheduling

Stateful conversational agent using LangGraph with conditional tool routing and live web retrieval.

🎯 What it does:

• Maintains multi-turn dialogue state across long conversations
• Automatically triggers web search only when queries exceed internal LLM knowledge
• Recovers gracefully from node failures without breaking conversation flow

🛠️ Tech Stack:

• Orchestration: LangGraph StateGraph, LangChain
• Search: Tavily Search API
• Frontend: Streamlit

✨ Key Features:

• 🔄 Stateful multi-turn dialogue with automatic state recovery
• 🌐 Conditional web retrieval → out-of-knowledge failures near 0%
• 🛠️ Tool routing with conditional edge logic

Production RAG pipeline for semantically accurate document Q&A with sub-second retrieval.

🎯 What it does:

• Indexes any PDF into a FAISS vector store and answers questions with source grounding
• Profiled and optimized with LangSmith tracing for production latency

🛠️ Tech Stack:

• RAG: LangChain, FAISS, OpenAI API
• Observability: LangSmith

✨ Key Features:

• ⚡ Sub-second nearest-neighbor retrieval via FAISS vector indexing
• 📉 150ms latency reduction (20% improvement) by refactoring to parallel embedding inference
• 🔍 LangSmith tracing for bottleneck identification and optimization

A comprehensive full-stack platform that revolutionizes business connections using AI and location intelligence.

🎯 What it does:

• Connects property owners, franchise companies, and entrepreneurs using AI algorithms
• Provides AI-powered market analysis and intelligent pricing recommendations
• Features modern responsive web interface with real-time AI insights

🛠️ Tech Stack:

• Backend: Python, FastAPI, SQLite
• AI/ML: Mistral AI for intelligent recommendations and NLP
• APIs: Foursquare Places API for location intelligence
• Frontend: HTML5, CSS3, JavaScript, Bootstrap 5

✨ Key AI/ML Features:

• 🗺️ Location Intelligence powered by Foursquare API
• 🤖 AI-powered business matchmaking using advanced algorithms
• 📊 ML-based market analysis and pricing insights
• 🔐 Smart API management with AI-powered validation

Machine learning-based system for detecting malicious URLs using advanced feature extraction and classification.

• AI/ML Tech: Supervised Learning, Feature Engineering, Classification Algorithms
• ML Models: Logistic Regression, Random Forest, XGBoost
• Features: Intelligent URL analysis, ML-powered threat detection, automated feature extraction

Comprehensive deep learning research analyzing different optimization algorithms on CIFAR-10 dataset.

• AI/ML Tech: Deep Neural Networks, Optimization Algorithms, Performance Analysis
• ML Models: Custom ANNs with SGD, Adam, RMSprop, Momentum optimizers
• Features: Performance visualization, convergence analysis, model comparison

Real-time digital signal processing with embedded C++.

• Tech: C++, Arduino, Embedded Systems, Signal Processing
• Features: Real-time signal processing, interrupt handling, serial communication

• 🥇 Bajaj Finserv HackRx 6.0 — Ranked 87th out of 7,000+ (Top 1.3%) for fintech solution design · Aug 2025

• 🧩 Going deeper into LLM evaluation frameworks and agentic system design
• ⚡ Exploring DSPy, Langfuse, and MLFlow for production ML observability
• 🔬 Building on my BiLSTM research — interested in VLMs and multimodal systems
• 🚀 Contributing to open-source AI projects in AgTech and health tech
• 🌱 Mastering MLOps and AI deployment at scale