02 — Projects
What I've Shipped
// PROJECT_01 · LangGraph & PostgreSQL
🧠
Agentic Text-to-SQL Engine with Autonomous Self-Correction
Python · LangGraph · PostgreSQL · Streamlit · FastAPI
▸Challenge: Naive text-to-SQL wrappers crash on complex queries — hallucinated schemas, syntax errors, and broken joins make them unreliable for production BI.
▸Solution: Engineered a stateful AI agent using LangGraph with an autonomous self-correction loop that parses PostgreSQL errors to dynamically rewrite and retry failed queries (up to 3 retries). Secured with database-layer Row-Level Security (RLS) for multi-tenant access control, exposed via REST API & Streamlit dashboard.
✅ 100% syntax validity before returning data to users
// PROJECT_02 · Enterprise AI Security
🛡️
Enterprise AI Gateway & PII Privacy Proxy
Python · FastAPI · Hugging Face (RoBERTa / Presidio) · Docker · Redis
▸Challenge: "Shadow AI" data leakage — employees paste proprietary code, customer PII, and financial data into cloud LLMs, violating GDPR and HIPAA compliance.
▸Solution: Built an AI middleware proxy that intercepts prompts via a local NER pipeline (RoBERTa/Presidio), masks PII before it leaves the network, sends sanitized prompts to the cloud LLM, then re-injects original data into the response — enabling complex reasoning without ever exposing protected data.
🔒 Zero PII exposure to external LLM APIs (HIPAA/GDPR compliant)
// PROJECT_03 · Semantic Search & ML
📄
Real-Time Research Paper Recommendation Engine
Python · Streamlit · FAISS · Sentence Transformers · TensorFlow
▸Challenge: Researchers waste hours sifting through thousands of arXiv papers to find relevant work — keyword search misses semantically similar papers with different terminology.
▸Solution: Built a real-time recommendation engine that indexes arXiv papers using dense embeddings (Sentence Transformers) into a FAISS vector store for instant semantic similarity search. Users search by paper title or abstract and get ranked recommendations in milliseconds via a Streamlit interface.
⚡ Sub-second semantic search across thousands of research papers
// PROJECT_04 · Chrome Extension & DOM Scraping
💬
AI Chat Exporter — Universal Chrome Extension
JavaScript · Chrome Manifest V3 · HTML · CSS
▸Challenge: AI chatbot conversations are trapped inside browser tabs — no native export, no way to archive, search, or feed prior context into another model.
▸Solution: Built a zero-dependency Chrome Extension (Manifest V3) that scrapes conversations from 14 AI platforms (ChatGPT, Gemini, Claude, Copilot, etc.) into clean Markdown or plain text files. Runs 100% locally — no API keys, no backend, no data leaves the browser.
🌐 14 AI platforms supported with one-click export
// PROJECT_05 · LangChain & Email Automation
📊
AI-Powered Job Application Command Center
Python · Flask · LangChain · Google Gemini · SQLite · Telegram API
▸Challenge: Job seekers lose track of applications buried across hundreds of emails — manually updating spreadsheets is error-prone and time-consuming.
▸Solution: Built a localized AI system that automatically ingests inbox emails via IMAP, processes them through Google Gemini (via LangChain) to extract company names, roles, and statuses, then stores everything in a local SQLite database. A Flask web dashboard provides a unified command center with optional Telegram alerts for real-time updates.
📬 Fully automated email-to-dashboard pipeline with AI extraction
// PROJECT_06 · Distributed Streaming RAG
🌊
Project Occam: Semantic Deduplication Pipeline
Apache Flink · Kafka · ONNX · Milvus · PyArrow
▸Challenge: Incident streams contain 85%+ redundancy. Routing this firehose directly to a cloud embedding API or vector database wastes money and massively bloats storage.
▸Solution: Engineered a distributed semantic gatekeeper using PyFlink and RocksDB. A local ONNX embedding model with SimHash LSH bucketing drops near-duplicate logs in real-time before they reach the cloud, drastically reducing API spend while preserving novel semantic intelligence.
📉 Cuts embedding spend and Milvus storage by dropping 85%+ redundancy