DEV Community: Juanita

RxQuery.AI: Your Dr. Internet Replacement powered by MindsDB!

Juanita — Mon, 30 Jun 2025 08:50:34 +0000

A pharmacist-AI that's here to replace your endless scrolling of symptoms and remedies on google search! Ask your queries and get smarter, faster and reliable responses.

Well, What is RxQuery and Why?

Imagine a world where AI acts like your own personal pharmacist, answering your drug-related queries, recommending safe alternatives, and analyzing side effects—all in seconds. Yea, that's our goal lol. RxQuery is an intelligent, AI-powered drug information finder and medicine assistant (think of a pharmacy simulator?) that provides instant, reliable drug consultation through an intuitive interface.

⚠️ Note: RxQuery.AI is not a substitute for professional medical advice. Please consult actual doctors, thanks :)

💡 Why We Built RxQuery.AI

With the vast amount of drug data available out there, we wanted to simplify access using natural language queries. Traditional search doesn’t cut it, semantic understanding is key ( basically understanding user context ). This allows users to narrow down their drug queries and get smarter and safer responses. To implement this semantic understanding, we use MindsDB's Knowledge bases.

✨ Key Features

General Purpose Assistant: RxAssistant (via AI Table)
AI Agents:

🔍 Drug Classification - Instantly classify medications (Antibiotic, Analgesic, etc.)
💊 Smart Recommendations - Get personalized drug suggestions based on symptoms
⚠️ Side Effects Checker - Comprehensive side effects analysis
🛡️ Allergy-Safe Search - Find safe alternatives for patients with allergies Command based input:
Command-based Interaction - We use simple slash commands like /classify, /recommend, etc to select the agent and input in a singular input field.
Voice Interaction - Talk to RxQuery for hands-free health queries
Intuitive UI/UX - Chat-based interface with real-time updates

🛠️ Tech Stack

Frontend: Next.js 14, TypeScript, Tailwind CSS, Framer Motion, Shadcn, MVPBlocks
Backend: FastAPI, Python, Pydantic

#🧠 MindsDB

CREATE KNOWLEDGE_BASE, INSERT INTO, CREATE INDEX(chromadb)
CHAINED MULTI AGENTS using CREATE AGENT for each feature (/classify, /recommend, /side-effects, etc.)
EVALUATE KNOWLEDGE_BASE with Groq for document scoring
CREATE JOB to ingest drug data periodically
SELECT ... WHERE content LIKE in semanticss!
metadata_columns to enable hybrid semantic + SQL filtering
CREATE MODEL rx_assistant with OpenAI for reasoning and classification
KB_EVALUATE: Groq LLM, AI TABLES: OpenAI, AGENTS: OpenAI, Ollama (experimental, model removed)
🧪 Editor: Our MindsDB SQL Editor code is included as reference for building/debugging Agents and KB queries.

P.S KNOWLEDGE BASES, AGENTS AND AI TABLES powered by MindsDB!

ARCHITECTURE!

Deep dive into implementation~

1.🧩 Knowledge Base Creation

Everything starts with the creation of a Knowledge Base (drug_kb), which acts as the centralized, semantically searchable store for drug data.

CREATE KNOWLEDGE_BASE drug_kb USING embedding_model = { "provider": "", "model_name": "", "base_url": "" }, reranking_model = { "provider": "", "model_name": "", "base_url": "" }, metadata_columns = ['category', 'usage'], content_columns = ['description'], id_column = 'drug_name';

metadata_columns: Supports hybrid queries (e.g., filter by category).
content_columns: The KB "reads" from description, basically here only we have the semantic matching.

2.📥Ingesting Data into the KB

Once the KB is created, we ingest structured data from a CSV file (medicine_details) into it

INSERT INTO drug_kb (drug_name, description, category, usage) SELECT drug_name, description, category, usage FROM files.medicine_details LIMIT 50;

We will be automating this ingestion later using Jobs

3.🔍 Semantic Querying

Now that our KB is populated, we can use natural-language-style semantic queries — thanks to the embeddings under the hood.

SELECT * FROM drug_kb WHERE content LIKE 'what drug to use for Fever and Headaches category?';
This gives us relevant drugs, even if the query doesn’t match word-for-word. Basically, semantics lol.

4.🧠 Chained Multi-Agent Pipeline

We’ve created a modular AI agent chain, where each step enriches context for the next. These agents access the KB for tailored outputs!

a. Classify Agent

CREATE AGENT classify_agent USING input_column = 'question', output_column = 'response', prompt_template = 'Classify the query "{{question}}" into a drug category like Antibiotic, Antipyretic etc.';

Classify agent obviously, used to infer drug class from user symptoms. (e.g., "fever" → Antipyretic).

b. Recommender Agent

CREATE AGENT drug_recommender USING input_column = 'question', output_column = 'recommendation', metadata_columns = ['category'], prompt_template = 'Based on drugs in category "{{category}}", what should user take for: {{question}}?';

This agent uses the output from the Classifier and the category metadata in the KB.

c. Side Effects Agent

CREATE AGENT side_effect_agent USING input_column = 'recommendation', output_column = 'side_effects', prompt_template = 'what are the common side effects of the drug "{{recommendation}}" ?';

Gives post-recommendation validation like on side effects and stuff, ensuring user safety.

d. Allergy-Safe Agent

CREATE AGENT allergy_safe_recommender USING input_column = 'allergy', output_column = 'safe_drug', prompt_template = 'Given the allergy: "{{allergy}}", recommend a safe drug that avoids triggering it. Also explain why it is suitable shortly.';

Well as name says, it filters recommendations based on allergy risk.

5. 🛠️ AI Tables: rx_assistant

In addition to agents, we used AI Tables for general-purpose assistance.

CREATE MODEL rx_assistant PREDICT response USING engine = 'openai', model_name = '', api_key = '', prompt_template = 'You are a helpful drug information assistant. If a user inputs a user query "{{question}}" and allergy "{{allergy}}", return helpful medicine suggestions...'
This model takes free-form medical questions and provides safe suggestions, optionally filtering with user allergy input.

6.⏱️Automated Ingestion: JOBS

CREATE JOB drug_kb_updater AS ( INSERT INTO drug_kb (drug_name, description, category, usage) SELECT drug_name, description, category, usage FROM files.medicine_details WHERE id > COALESCE(LAST, 0) ) EVERY 1 hour;

Well this just makes sure everything is upto date and kb is updated...this occurs every 1 hour.

7.⚠️ We do our knowledge base evaluation using Groq with the command EVALUATE KNOWLEDGE_BASE

✨ ✨ To the curious folks,

If y'all are interested and wanna try it out, you can do so on your local:

Prerequisites

Python 3.8+
Node.js 18+

Backend Setup

cd backend pip install -r requirements.txt uvicorn main:app --reload --port 8000

Frontend Setup

cd frontend/rxquery npm install npm run dev

Something like this~

Command	Example	Result
/classify	/classify paracetamol	→ "Antipyretic"
/recommend	/recommend headache	→ "Paracetamol 500mg every 6 hours"
/side-effects	/side-effects Ibuprofen	→ "Stomach upset, kidney issues with long-term use"
/allergy	/allergy penicillin	→ "Try Azithromycin (macrolide class)"
/general	/general tired and dizzy	→ *"Drink warm water, eat light foods and take dolo."

🍒 Acknowledgments

MindsDB team for the amazing AI platform
Kaggle for the dataset
Open source community for inspiration

Thank you for reading, hope you all like it and try it out! :)

Building Forger AI: Craft Cold-Emails with the Precision of Yor!

Juanita — Mon, 30 Sep 2024 07:54:21 +0000

Introduction ✨

Let's be honest-crafting personalized cold emails is NO FUN. I mean, we all know we should customize each mail, but who’s got time for that? That's the motivation behind Forger AI.

Now, What is Forger AI?

Well, It's a tool I created to automate the art of cold emailing, making you sound like you put effort into it (even when you didn’t). In this blog, I’ll break down how I tried out multiple APIs, ran into more issues than I thought I would, and somehow ended up with a solution that works lol.

Checkout the demo here: Youtube Demo Video

The Motivation behind Forger AI 🔥

Why did I build this? Simple—I hate manually writing cold emails. With Forger AI, the idea was to automate email generation in a way that felt personalized and human. The goal? Paste a job post URL, and yayy!—AI whips up a well-crafted email that is relevant.

But, of course, I couldn't just stop at generating one email at a time. Why not send hundreds in bulk while I’m at it? This is where APIs like Llama3.1 and SendGrid API come into play.

The Tech Stack:

Frontend: Built with Next.js and TypeScript for a clean, scalable interface.
Tailwind CSS: For styling the interface
Backend: FastAPI paired with Langchain to handle the logic and API routing.
Llama API: The heavy lifter for parsing job descriptions and crafting custom email content.
SendGrid API: Sends all those beautifully generated emails into the void (aka recruiter inboxes).
Stripe API: Experimented with adding a payment gateway for premium features, and Stripe was the go-to solution for handling smooth payments.
GroqCloud: Manages CSV uploads for bulk recruiter emails

Architecture Diagram

Play around with the code here on my Notebook: email_generator.ipynb

✨ Features:

Cold Email Generation

You can generate humanish-personalized cold emails from job post URLs.
Just enter a URL and AI will take care of the rest lol
Send Emails to Recipients

Send the generated emails directly to the recipient’s inbox. No need of
Copy-Pasting.
Upload Portfolio CSV

Upload a CSV file with your portfolio and attach it to your emails
automatically. Saves time!
Bulk Recruiter Upload
Upload a CSV file with recruiter emails, and we’ll send your cold
emails to all of them at once.
Premium Features

Unlock extra features like bulk email sending and advanced
customizations with a premium upgrade.
Edit Email (Coming Soon!)

Currently working on editing the AI-generated emails before sending.
Email Preview

See a live preview of the email generated by AI before sending the
mail.
Copy Email to Clipboard

Copy the generated email with one click and use it wherever you want.

THE API PHASE!🔥

This phase involved a lot of experimentation and broke a few things before finally selecting one.

I tried out OpenAI models ( GPT-3/GPT-4 ) and though it was super-versatile, it's pretty generic and the cost of using GPT-4 for large-scale tasks adds up pretty quickly. Then we went on with Claude and it was also quite similar.

API/Model	Pros	Cons	Why Not? / Why Chosen
GPT-3 / GPT-4 (OpenAI)	- It's obviously versatile, powerful & handles a wide range of tasks. - Well-established ecosystem and integrations. - Generates large, cohesive content.	- But it's too formal and generic. - Too $$$ for large-scale use lol. - Overkill for specific cold emailing tasks.	- Emails felt too “corporate” and lacked the human-ness?. - I was looking for something more focused for job-specific cold emails.
LLaMA API	- Context-aware, produces personalized humanish emails. - Natural, conversational tone. - Requires less fine-tuning. - Fast and efficient performance.	- Newer ecosystem, so lacks the broader integrations. - Slightly less community support compared to GPT though.	- Finally chose this for efficient, human-like, job-specific email generation without the need for excessive fine-tuning.
Claude (Anthropic)	- Ethical AI design, prioritizes safety. - Great for general-purpose tasks, such as conversation and customer service. - Fast response times.	- Lacked the customization needed for cold emails. - Less flexibility in fine-tuning for specific, tailored content. - More focused on broader use cases.	- Didn’t meet the personalized email requirement and wasn’t as adaptable to job-specific cold emails.
GroqCloud API	- High-performance AI infrastructure. - Optimized for large-scale tasks and ML model deployment. - Efficient use of resources for certain operations.	- Overkill for this project’s focus. - More suited for ML tasks rather than personalized email generation. - Complex setup.	- Too complex for the goal of generating cold emails. - More focused on infrastructure and ML tasks, which wasn’t necessary for this project.

Future plans for Forger.AI:

For now, I'm working on adding AI Agents that can handle email-follow ups and responses which AGAIN, can aid my laziness lol. Also, right now, Forger AI personalizes emails based on job postings. Next up, I'll be integrating LinkedIn and other sources to craft emails tailored to individual profiles. Currently improving the premium feature ie, the bulk recruiters email data for automating email sending process.

Final Thoughts <3

Building Forger AI was a journey of many APIs, lots of coffee, and a few facepalm moments. Well atlast, we made it though, a solution to automate the boring cold-email content process. There's still more to come though, more features and more learning!

If you read till here, thank you so much and I hope it was interesting!

Contact me 😎

Email: juanitacathy486@gmail.com
Github: JuanitaCathy

Hacktoberfest 2023- Learnings and Experience!

Juanita — Mon, 30 Oct 2023 16:45:55 +0000

Introduction

I'm Juanita, a passionate undergrad who is exploring open-source and learning while doing. This was my second Hacktoberfest and was filled with learnings. Got to work on some amazing projects and learnt something new which was fun!
You can checkout my Github here -> https://github.com/JuanitaCathy

Journey Throughout

This Hacktoberfest was certainly filled with many highs and lows, one of the lows being me struggling to find a repository to contribute to. Initially, as I wasn't well aware of the trending tech, I found it hard to get hold of a repository to contribute to. However, when I found a repository which somewhat aligned to what I wanted to do, I went through their codebase and learnt some basic stuff and contributed with what I learnt.

I was learning by doing and this was helpful as it wasn't just a one-time thing and will help me in a long run. I started with basic contributions like python scripts and went on to go abit advance. This was a massive boost in confidence as well which is nice. It was a nice experience and the maintainers of the repository I contributed to were also pretty chill which further enhanced the experience.

Growth

When I started out with Hacktoberfest, I wasn't sure if I could apply my theoretical knowledge of a specific language for real-world applications. However, with me learning things on the go and applying them made me more confident and strong with my foundations.

I initially knew only Python, however, there was this one contribution where I had to add Poetry support and that was a really big confidence boost once I finished as when I started out, I had no clue what Poetry was. For now, I'm having fun learning and creating an impact through open-source. Hacktoberfest 2023 really is my inflection point and I love it.