نبذة عني
AI Engineer who builds and ships LLM-powered systems that real teams use every day. I specialize in agentic AI, retrieval-augmented generation (RAG) and backend services, and I take projects from first prototype to produ…
AI Engineer who builds and ships LLM-powered systems that real teams use every day. I specialize in agentic AI, retrieval-augmented generation (RAG) and backend services, and I take projects from first prototype to production deployment.
At Extreme Branding, an Amazon marketing and PPC agency, I designed and built an AI coworker that works inside the company’s Slack. Staff can ask questions, upload files and get answers in the same thread, with context drawn from recent channels and messages. The system uses tool calling to assign Asana tasks, follow up with team members and report back to task owners. It connects with Slack, Asana, Gmail, Google Drive and WhatsApp through a FastAPI backend with authentication, workspace-scoped data and automated tests. To keep costs low, I route requests across models: lightweight models for quick replies, stronger ones for analysis.
Outside of work, I have built and published several production-style projects. My SQL Analyst Agent is a multi-agent LangGraph system that turns plain-English questions into safe SQL. It uses an AST-based safety guard, human approval for risky queries and a self-correction loop. It scored 100% execution accuracy (lenient) on a 40-question benchmark, blocked 5 out of 5 malicious prompts, and runs 176 tests in CI. I am also distilling text-to-SQL ability from large models into a 1.5B model that runs on a laptop, using LoRA fine-tuning and DPO.
My background in psychology (BS, 3.87 CGPA) shapes how I evaluate AI systems. I have published research repositories that test LLMs on cognitive psychology tasks and decode brain signals from EEG data using deep learning.
Core skills: Python, FastAPI, LangGraph, LangChain, LangSmith, RAG, vector databases (ChromaDB), OpenAI, Gemini, Groq, Ollama, Docker, GitHub Actions, SQL and Scikit-learn.
الخبرة
Junior AI Engineer
<p>Built and deployed Tally, an LLM agent in Slack that answers from company documents, Slack, Asana and Gmail/Drive with source citations and per-user access control; FastAPI, Next.js, SQLite, Docker, DigitalOcean.<br>
Raised answer accuracy on a 14-question benchmark from 71% to 93% (facts 30 → 34 of 36) by redesigning the context pipeline of the tool-calling agent, and added a grounding checker that brought misattributed names to zero in evaluation.<br>
Lifted task-status accuracy from 55% to 95% (11 → 19 of 20 test cases) by combining LLM event extraction with a deterministic state engine, at $0.004 per evaluation run.<br>
Automated Asana task assignment and deadline reminders: assignee ranking by past work and workload, working-hours scheduling, interactive Slack buttons and per-user OAuth sign-in, with human approval before every write action.<br>
Integrated WhatsApp through headless Chromium in Docker; traced a chain of silent failures in the open-source whatsapp-web.js library and took message ingestion from zero to working in production.<br>
Kept the system reliable and cheap: ~900 automated tests, 25 backward-compatible schema migrations, multi-tenant workspace isolation, and model routing with daily and lifetime spend caps.</p>
Freelance Data Analyst & Research Assistant
<p>Cleaned, analysed and reported research datasets with Python, Pandas, Excel and SPSS; wrote scripts that automated recurring data processing and cut turnaround time.</p>
Artificial Intelligence Intern
<p>Built conversational chatbots and NLP text-analysis pipelines using TF-IDF, Word2Vec and Bag-of-Words.<br>
Contributed to an Intelligent Document Processing system that extracts and validates fields from invoices and PDFs.<br>
Evaluated LLM applications for accuracy and efficiency on real-world scenarios.</p>
المشاريع
AI Coworker for Slack (Agentic Workplace Assistant)
<ul><li>Project Title: AI Coworker for Slack (Agentic Workplace Assistant)</li><li>Organization Name: Extreme Branding</li><li>Project URL: leave blank (this is internal company work)<ul><li>Project Description:Built an AI coworker that works inside Slack: staff ask questions, upload files and get answers in the same thread, using recent channel and message context.</li><li>Designed tool calling for Asana: the assistant picks the best assignee, creates the task, follows up and reports back to the owner.</li><li>Built the FastAPI backend with login, workspace-scoped data and automated tests; integrated Slack, Asana, Gmail and Google Drive.</li><li>Cut AI cost by routing requests across models (GPT-4o mini, GPT-5 mini, Gemini) with daily usage limits.</li><li>Tools: Python, FastAPI, Next.js, OpenAI, Gemini, Slack API, Asana API.</li></ul></li></ul><p></p>