AI Engineer building production-ready LLM applications: Retrieval-Augmented Generation systems, tool-enabled AI agents, and guardrailed NL-to-SQL pipelines.
Focused on modular AI architecture, safety guardrails and refusal logic, vector database integration, and scalable backend services with FastAPI and PostgreSQL.
E-Commerce Intelligence Agent, Guardrailed NL-to-SQL BI Agent (June - July 2026) LangGraph-orchestrated agent answering free-form business questions against a live Postgres database via a schema-agnostic NL-to-SQL pipeline. Multi-layer SQL safety guardrails (AST validation, PII denylist, cost guard, read-only DB role), live production deployment on Railway and Vercel with an SSE-streamed chat UI and MCP server, evaluated at 90% execution-accuracy on a standard text-to-SQL benchmark. Tech stack: Python, LangGraph, LangChain, FastAPI, PostgreSQL, sqlglot, Next.js, TypeScript, MCP, LangSmith, Promptfoo, Docker, Railway, Vercel
Tool-Enabled AI Agent System (February 2026) Production-ready AI agent that decides whether to call external tools, executes them, evaluates safety via guardrails, and returns structured responses, with a modular tool registry, structured JSON logging, and deterministic decision logic. Tech stack: Python, FastAPI, PostgreSQL, Pydantic, Ollama, PyTest
Customer Support Knowledge RAG System (January 2026) RAG system built with a 4-person team during the Jatis Mobile internship, for customer support queries on pricing, subscriptions, and product features. Owned the Confidence Scoring and Refusal Logic module, using NLI models to detect contradictions across retrieved contexts and rule-based refusal conditions to reduce hallucinated answers. Tech stack: Python, FastAPI, PostgreSQL, pgvector, SentenceTransformers, Transformers, Ollama, PyTorch
Production-Ready RAG System (December 2025) Modular RAG pipeline supporting multi-format documents (text, PDF, tables, OCR images), with ingestion, chunking, embeddings, and pgvector similarity search, plus prompt-injection safeguards and fallback handling. Deployed via FastAPI. Tech stack: Python, FastAPI, PostgreSQL, pgvector, SentenceTransformers, Google Gemini, Ollama, Tesseract OCR
BPJSChaBo, BPJS Chatbot with NLP and RAG Pipeline (July 2025) AI-powered chatbot answering BPJS-related questions via RAG and NLP. Built the data engineering side: collection, cleaning, validation, and MongoDB vector storage, deployed on Hugging Face with Streamlit. Tech stack: Python, Pandas, LangChain, Gemini Flash, Deepseek-V3/R1, MongoDB Atlas Vector Search, Streamlit
End-to-End Pipeline for E-Commerce Shopping Behavior Analytics (June 2025) Automated pipeline pulling raw shopping data from PostgreSQL, cleaning and validating with Pandas and Great Expectations, and pushing to Elasticsearch, visualized through Kibana dashboards on seasonal trends and discount impact. Tech stack: Python, PostgreSQL, Apache Airflow, Elasticsearch, Kibana, Great Expectations
More projects
Skin Type Classification using CNN (June 2025) CNN model classifying facial skin types, deployed via Streamlit. Python, TensorFlow, Keras, scikit-learn.
Heart Failure Mortality Prediction (June 2025) Supervised learning model predicting mortality risk from clinical data. Python, Scikit-learn, XGBoost.
Revenue Strategy Design for Road Bikes in Canada Data-driven analysis proposing a 35% revenue growth strategy, presented via Tableau. Python, Pandas, Tableau.
Building AI systems that know when to answer, and when to refuse.