⚡
Engineering
2026-02-18·6 min
What Building Real AI Systems Taught Me (That Courses Don't)
From models to production: lessons from shipping AI used by real users.
Read moreBlogs
Thoughts on building AI systems, research, and engineering.
📄
AI Systems
From RAG to Agentic AI: What Actually Changed?
Lessons from building real multi-agent AI systems and why the paradigm shift matters.
Coming Soon🔬
Research
Why LoRA & QLoRA Work So Well for Legal Summarization
Practical insights from fine-tuning LLMs for long-form legal documents at scale.
Coming SoonExpertise
What I Build
Agentic AI Systems
Multi-agent workflows using LangGraph, LangChain — with tool-calling, memory, planning, and state transitions.
AI Backend Engineering
Fast AI backends: async pipelines, WebSocket streams, GraphQL and REST APIs, real-time AI integrations.
Machine Learning Systems
End-to-end ML pipelines from data preparation to deployment, model monitoring, and retraining workflows.
Applied Research
Applied research in legal AI and healthcare — including fine-tuning BERT, LLaMA, and building custom evaluation workflows.
My Roadmap
From foundations to scale — a journey in building AI.
Phase #01 — completed
Foundation Building
Built strong foundations in machine learning, deep learning, and backend engineering through coursework, certifications, and hands-on projects. Focused on understanding core ML concepts while actively building.
Phase #02 — current
Systems & Research
Currently building production AI systems: agentic workflows, LLM pipelines, and applied research in legal AI and healthcare monitoring. Focused on shipping real products.
Phase #03 — next
Scale & Impact
Aim to scale AI systems, publish research, and contribute to open-source ecosystems. Building toward global AI engineering or applied research opportunities.
Battle Tested 🏆
Hackathons won, problems solved, systems shipped under pressure.
Winner
Build for Billions: IndiaStack Hackathon
NITK Surathkal
Cultural & Heritage Domain
SuperMuseum
Runner-Up
IEEE Xplore Hackathon 2025
Ra.AI
Runner-Up
IETE-SF ACE 2.0
Emotional Intelligence Domain
Ra.AI
2nd Runner-Up
Taqneeq CyberCypher 5.0
Advanced Domain
Lorri AI
Projects I've Worked On 🚀
Real systems, shipped to production.
Lorri AI
An agentic logistics control tower that autonomously consolidates freight shipments to reduce trips, cut costs, and lower emissions. Runs a full observe-reason-decide-act-learn loop with a domain-specific ML model, OR-Tools solver, policy guardrails, and 10-node LangGraph state graph. 70% trip reduction, 41% cost savings, full pipeline in under 1 second.
FastAPILangGraphscikit-learnOR-ToolsGemini 2.0ReactDocker
Serene.AI
Emotional wellness multi-agent system built with LangGraph and dynamic state transitions across Insight, Context, Sentiment, and Crisis modes. Features hybrid RAG (FAISS + BM25), triple memory stack with MongoDB, and achieves 1.2–1.5s latency under emotionally complex multi-turn dialogue.
LangGraphFAISSMongoDBFastAPIReact
🔍
SuperHoax.AI
Agentic misinformation detection system that identifies and verifies viral claims during global crises. Specialized AI agents monitor social media and news in real time, cross-checking claims against credible sources. Builds dynamic reputation scores for information sources, enabling faster trust decisions at scale.
FastAPILangGraphGeminiTavily SearchReactDocker
Private repository
🏛️
SuperMuseum — AI Cultural Guide
AI-powered interactive digital museum with a gamified isometric pixel-art environment and multilingual agentic RAG guide. Users explore Indian heritage through personalized tours in 10+ languages. Winner of the Experience India Track at IndiaStack Build for Billions Hackathon at NITK Surathkal.
LangChainLangGraphFastAPISarvam AIJioSaavn API
💬
GDG WhatsApp AI Chatbot
AI-powered WhatsApp assistant for GDG on Campus NMIMS that helps students discover events, explore technical domains, and learn about recruitment. Intent classification pipeline with Gemini-powered LLM, conversation state management, and engagement analytics. 100+ students interacted during Foobar outreach.
FastAPIGemini AILangChainGo WhatsApp APIDocker
Private repository
📞
Payment Reminder Caller Agent
AI-powered system that automatically places multilingual voice calls (Hindi, English, Marathi) to customers with pending invoices, reducing manual follow-up by 80%. Real-time WebSocket audio streaming via Twilio and Vapi AI, AI-driven transcript analysis for intent and sentiment, and async FastAPI backend with APScheduler for daily automation.
FastAPITwilioVapi AIWebSocketGoogle Sheets API
Private repository
Get In Touch
Let's build something incredible together.