I’m a 2024 graduate from IIT Kanpur with a background in Physics, currently focused on building production-grade AI systems across:
- Generative AI
- Agentic workflows
- Retrieval-Augmented Generation (RAG)
- Applied Machine Learning
- Backend APIs and deployment
I like working on problems where the challenge is not just model performance, but how the whole system behaves end to end reliability, control flow, retrieval quality, deployment, and observability.
|
|
|
|
Production-style AML investigation system combining classical ML with LLM-based agentic reasoning.
What it does
- Detects suspicious patterns in financial transactions
- Uses LLM agents to investigate flagged cases
- Generates audit-ready reports for review workflows
- Adds privacy and explainability layers for safer deployment
Key highlights
- Built on 9.5M+ transaction records
- Used CatBoost for highly imbalanced fraud detection
- Designed LangGraph workflows for evidence gathering and report generation
- Deployed with FastAPI, Docker, and AWS
- Added SHAP for model explainability
Tech: Python CatBoost LangGraph Llama FastAPI Docker AWS SHAP
A retrieval system that lets users query information across PDFs, web pages, DOCX files, and YouTube transcripts.
What it does
- Ingests multiple unstructured data formats
- Creates semantic retrieval pipelines using embeddings
- Returns grounded answers with citations
- Supports conversational querying through a UI
Key highlights
- Built with Sentence-Transformers + ChromaDB
- Used LangChain and Gemini
- Reduced hallucinations using retrieval grounding
- Added conversational memory and interactive querying
Tech: Python LangChain Gemini Sentence-Transformers ChromaDB Streamlit
A closed-loop LLM-powered debugging workflow that detects, patches, and validates Python code issues.
What it does
- Runs buggy code in a controlled environment
- Detects syntax, runtime, and logic errors
- Generates patches automatically
- Re-validates code after fixes
Key highlights
- Designed a loop: Execute → Parse → Patch → Validate
- Added sandboxed execution for reproducibility
- Built a debugging UI with logs and diff visualization
Tech: Python Gemini Docker Streamlit Agentic Workflows
- Top 10 Finalist SARCathon, PAN-IIT AI Hackathon, IIT Bombay
- Built a multilingual RAG-based FAQ system under time-constrained hackathon settings
- Interested in building AI systems that balance speed, reliability, and practical utility
- Reliable agentic AI systems
- Better retrieval quality for RAG
- Production-ready LLM application design
- AI systems that are observable, controllable, and deployable
- Email: karthikkalikivayi2206@gmail.com
- LinkedIn: linkedin.com/in/karthik-kalikivayi-34a669215
- GitHub: github.com/The-Name-is-Karthik