Basil Latif's AI Portfolio

Selected AI projects, from client engagements to personal builds:

Project Name	Affiliation	Role
Prescription SIG Extraction	Client Engagement: Pharmacy Software Company	Built a clinical NLP system that turns free-text prescription directions into structured, validated data across 100K+ prescriptions
AI Legal Skills Platform (Litigation Practice)	Client Engagement: Solo Litigation Practice	Led long-record review and legal research skills on a Claude-based platform, with source-cited summaries, a citation verification gate, and an eval suite built from redacted case materials
Fully Local AI Analyst Agent	Personal Project	Built a fully local AI agent using Qwen, FAISS, and RAG to answer questions about personal datasets with complete data privacy, documented on Medium
Vector Database Standup with Pinecone	Professor at LMU	Used a vector database to store e-commerce product data and then used it to retrieve similar products
Practical Multi AI Agents and Advanced Use Cases with crewAI	DeepLearning.Ai	Deployed agentic crews using CrewAI
Retrieval Augmented Generation (RAG) Demo	Presentation to Great Learning	Explained a business use case for RAG and showed a HuggingFace demo
LangChain Chat with Your Data	DeepLearning.ai Course	Did projects to understand how to connect an AI agent to a database
RAG Chatbot: Data Automation with Chatbot for Housing Analytics	GWU MSBA Capstone Project	Built a chatbot that queries Snowflake data warehouse for real-time housing analytics answers
LLM Email Automation System	Personal Project	Deployed an email drafting application using LLM technology, documented on Medium

Project Details

Client Engagement · Pharmacy Software Company

Prescription SIG Extraction

A clinical NLP system that converts free-text prescription directions into structured data.

Prescription directions (SIGs) arrive as messy, abbreviation-heavy text like 1 tab PO BID PRN. This system turns them into structured fields: dose, form, route, frequency, duration, and special instructions.

Language model: a 20B-parameter model interprets the free-text input.
Validation: a deterministic Python parser checks the output and enforces a consistent schema, so downstream pharmacy systems receive clean, reliable fields.
Scale: runs on 100K+ prescriptions with sub-second inference per SIG.

Client Engagement · Solo Litigation Practice

AI Legal Skills Platform

An AI-powered legal skills platform built on Claude for a solo litigation attorney with a heavy active caseload.

The platform is organized around five skills. I led two of them and built the evaluation that tests them:

Long-record review and summarization: page- and section-aware chunking so every summary point traces back to a source citation.
Legal research: a mandatory verification gate before any citation reaches the attorney.
Evaluation: an eval suite built from redacted case materials that tests both skills against actual filings and hearing prep, not synthetic examples.

Personal Project · Documented on Medium

I Built a Fully Local AI Agent That Works Like an Analyst

A fully local AI agent using Qwen (via Ollama), FAISS, and a RAG pipeline to answer questions about personal datasets with complete data privacy.

Read the Article on Medium

GWU MSBA Capstone · Team Supervisor

RAG Chatbot: Data Automation for Housing Analytics

A chatbot that queries a Snowflake data warehouse to produce real-time answers to natural language questions.

I supervised a team of Master's students to produce this deliverable.

Housing analytics RAG chatbot screenshot

Personal Project · Documented on Medium

I Built an LLM That Writes Emails for Me

An LLM-powered application that drafts emails in my voice, with a walkthrough of how it was built.

Read the Article on Medium