I build AI agents and RAG systems, from retrieval to evaluation.
Hi, I'm Aditi. I spent nearly 8 years in analytics and software engineering across healthcare and banking, working with sales, prescription, and patient data. Now I build agentic AI systems with human review, audit trails, and evaluation built in.
M.S. in Computer Science, Stevens Institute of Technology.
Selected Projects
Agentic AI systems, RAG and evaluation, ML apps, and fine-tuning work.
Patient Risk Flagging Agent
A LangGraph agent that reads a clinical note, checks medications against kidney function, and scores patient risk as Low, Medium, or High. It pulls clinical guidance with RAG (ChromaDB), pauses for human review on high-risk or unsafe cases, and logs every outcome to an audit trail.
View on GitHub ↗Prior Authorization Multi-Agent System
Four CrewAI agents (extractor, policy checker, writer, reviewer) that check an MRI request against a payer's criteria and draft the authorization letter. If the reviewer finds the letter overstates the case, it sends it back to the writer with feedback, up to three rounds.
View on GitHub ↗Browser Agent Reliability on Yelp
A browser agent that finds restaurants on Yelp, tested on 40 hand-scored runs. It claimed success every time but was wrong a third of the time. A stricter prompt made it 2.5 times cheaper but over-cautious; letting code make the final choice from the agent's own list reached 80% correct at 40% of the cost.
View on GitHub ↗RAG Chatbot and RAG Evaluation
A chatbot that answers questions from an uploaded PDF, and an evaluation pipeline that measures retrieval and answer quality.
Repo coming soonConversational Sales Data Agent
Ask a question about sales data (57K+ transactions) in plain English and get an answer. A LangGraph agent turns the question into SQL, runs it, and explains the result. Served through a FastAPI endpoint with caching, so repeat questions come back fast.
View on GitHub ↗Insurance Premium Category Predictor
An end-to-end ML app: a RandomForest model with engineered features (BMI, age group, lifestyle risk, city tier), served through FastAPI with Pydantic validation and a Streamlit front end. Packaged with Docker and published to Docker Hub.
Repo coming soonQLoRA Sentiment Classifier
Fine-tuned TinyLlama for sentiment classification using QLoRA (4-bit quantization plus LoRA) on a free Colab GPU.
View on GitHub ↗Technical Skills
What I build with most
Python, SQL, LangGraph, CrewAI, RAG, FastAPI, and Tableau.
Also work with
AI & Agents: LangChain, ChromaDB, RAG evaluation, QLoRA fine-tuning
Backend & Web: Pydantic, Docker, Streamlit, HTML, CSS, JavaScript, Node.js, React.js
Data & Analytics: MS Excel (Advanced), market mix modeling, segmentation, patient analytics
Machine Learning: scikit-learn, logistic regression, neural networks, probabilistic models, MLE, gradient ascent
Professional Experience
Business Analyst
Jul 2024 – Mar 2025Conversational AI agent for sales data questions
- Built a conversational AI agent with LangGraph, GPT-4o, FastAPI, and SQLite that turns plain-English questions into SQL over sales data (57K+ transactions), so sales reps can get answers on revenue, territory, and customer trends without waiting on the analytics team.
Commercial excellence support for an oncology product (client: Helsinn Therapeutics)
- Used fuzzy matching on account demographics to map OneOncology accounts to 867 sales data, then tracked volume shifts and each group's share of growth.
- Built volume concentration curves for Akynzeo to show how business was spread and find the top accounts.
- Built an interactive Tableau dashboard for Akynzeo covering key KPIs and sales trends.
- Delivered weekly rebate entity reports for OneOncology, Unity, and multiple territories with Excel macros, and prepared Akynzeo data from SQL databases for the KAM tool.
Associate
Sep 2020 – Dec 2021Market mix modeling for a heart failure product
- Estimated the sales driven by each promotional channel (calls, samples, emails, direct mail, paid search), accounting for brand equity, sales lag, and historical efforts.
- Modeled current-month prescriptions in Python as a function of promotion and previous-month prescriptions to capture lag effects.
- Segmented physicians by specialty, market decile, and NRx decile, and calculated optimal spend and ROI by channel and segment.
Senior Software Engineer
Apr 2017 – Sep 2020Commercial excellence support for a specialty migraine product
- Delivered weekly and monthly reports with the onshore team on KPIs such as referrals, active patients, new starts, and net sales.
- Created at-risk reports to flag patients likely to drop off therapy, and built Tableau dashboards for the sales and marketing teams.
Internet banking and T24 development for Nedbank Africa
- Built Temenos Transact (INFO BASIC) routines for Nedbank's corporate internet banking during its move to Oracle FLEXCUBE.
- Extracted account and loan data for credit bureau reporting, ran Close of Business (COB) in production, and fixed COB incidents.
Programmer Analyst
Jan 2015 – Apr 2017Development and customization for Bank Julius Baer, Singapore
- Developed and supported core banking applications on Temenos T24, including data customization, data migration, and integration.
Education & Continuous Learning
-
M.S. in Computer Science · GPA 3.93/4
Jan 2022 – Dec 2023Stevens Institute of Technology, Hoboken, NJ · Machine Learning, Mining Big Data, Cloud Computing, Database Management Systems. Built full-stack web apps with React, Node.js, MongoDB, and Redis.
-
B.Tech. in Computer Science · GPA 3.6/4
Jul 2010 – May 2014Rajasthan Technical University (SKIT), Jaipur, India
-
Probability for AI (PAI)
2026 · In progressStanford University · Code in Place · Six-week program on probability, statistical inference, maximum likelihood estimation, and Bayesian reasoning. View portfolio ↗
About Me
Aditi Gupta
AI Engineer · Applied AI
I've worked with pharmaceutical sales data and built enterprise banking software, so I know how messy real business data is and how much accuracy matters. Now I'm focused on AI engineering: building agents that work with that kind of data, with the checks that make them safe to trust, like human review, audit logging, and evaluation.
My path started in core banking software on Temenos T24, moved into commercial analytics for pharma products, and then to an M.S. in Computer Science at Stevens Institute of Technology.
- Education M.S. in Computer Science, Stevens Institute of Technology B.Tech. in Computer Science, Rajasthan Technical University
- Background Nearly 8 years in analytics and software engineering across healthcare and banking
- Learning Stanford Probability for AI (in progress, 2026)
- Based in Somerset, New Jersey, United States
- Open to AI Engineer, Applied AI, and Forward Deployed Engineer roles
Let's Connect
I'm open to full-time AI Engineer, Applied AI, and Forward Deployed Engineer roles, and I'm always happy to talk with people building AI for real business data.