Rishabh Sharma
a full-stack developer and AI product builder with deep experience across machine learning engineering, scalable web architectures, and performance driven system design.
a polymath who bridges neural network fundamentals with production deployment to create impactful, enterprise grade solutions.
Experience
I'm building the core infrastructure for an intelligent trading platform that translates natural-language inputs into structured trading logic and executable workflows, enabling traders to express complex strategies conversationally and have them executed with precision.
At the heart of my work is developing financial modeling pipelines and trading recommendation engines that process real-time market data and generate actionable insights using machine learning. I'm engineering systems that integrate live data streams, historical patterns, and risk assessment frameworks using Python, FastAPI, and PyTorch. These pipelines handle high-frequency data processing, news sentiment analysis, and alternative data feeds all optimized for low-latency execution to support production trading environments.
Actively building autonomous trading agents that combine NLP with financial domain expertise to understand trader queries, interpret market signals, and recommend optimal strategies. The agent architecture uses RAG and LangChain to enable reasoning about complex scenarios, real-time backtesting, and explainable recommendations. I'm designing multi-agent workflows where specialized agents market analysts, risk assessors, and execution planners collaborate to deliver comprehensive trading intelligence.
This role involves developing robust ETL pipelines, building secure API integrations with trading platforms for order execution and portfolio management, and creating tooling infrastructure that ensures regulatory compliance. I'm leveraging my full-stack expertise (MERN, FastAPI), AI/ML skills (PyTorch, LangChain, NLP, RAG), and systems optimization experience to deliver enterprise-grade trading automation.
Worked on developing an AI-powered nursing engine designed to enhance diagnostic accuracy and streamline clinical workflows for healthcare practitioners. My primary focus was building and optimizing generative AI models that could process medical literature and provide evidence-based recommendations to support nursing staff in real-time decision-making.
Leveraging high-performance GPU infrastructure (NVIDIA A100, V100, A4, T4) to accelerate model training pipelines, reducing training time by 40% and enabling rapid iterative improvements. Working with the PubMedQA dataset, I fine-tuned large language models to achieve 85.2% diagnostic accuracy, significantly improving the reliability of AI-assisted medical recommendations.
This involved implementing advanced NLP techniques, RAG architectures, and domain-specific prompt engineering to ensure the system could handle complex medical queries with precision. I seamlessly integrated the AI engine with existing nursing systems, resulting in a 30% reduction in time spent on routine tasks and allowing nurses to focus more on direct patient care.
Throughout this internship, I applied my expertise in LLMOps, PyTorch, LangChain, and NLP to build production-ready AI systems for healthcare and enterprise compliance. I gained hands-on experience with GPU optimization, model fine-tuning on domain-specific datasets, system integration, and building scalable ML pipelines that deliver measurable impact in real-world clinical and enterprise environments.
In Between These Experiences
I've been building and shipping products since my first year of college. Each one taught me something different about users, systems, performance, and what it actually takes to solve real problems at scale.
It started with CNN library in Java No external ML libraries, just pure implementation. I optimized it with multi-threading, hit 96.5% accuracy on MNIST, and achieved 3x faster execution than naive Python implementations. That project taught me fundamentals matter more than fancy frameworks.
Next came Clone Defender, built a real-time MERN application with over 90% accuracy, serving 50+ security analysts. This was where I learned scalable architectures Node.js with worker threads, async task handling, processing large APK files efficiently. The system reduced threat response time by 60%. Real users, real impact.
After that, I built Blackout, it is a data-first web application built to collect, analyze, and visualize incidents of censorship across India. The project was started during the SFLC.in Hackathon to provide a transparent public record of removals, blocking orders, and censorship decisions.
Most recently, I worked on Multimodal RAG System, a production style system that enables question answering over complex documents containing text, tables, charts, figures, and scanned images.
So yes, hard work and consistency pay off. Each product was a step forward, even when it didn't feel like it at the time.
Education
2023 - Surviving
Achievements
Project: BlackOut
Built during the hackathon, it is a data-driven platform that documents free speech violations in India, including content take downs, website blocking, and film censorship. It features a searchable archive, interactive India map, and dashboards that make censorship trends more transparent for journalists, researchers, and citizens.
Project: GeneHackAMR
A machine learning-powered tool designed to predict antimicrobial resistance (AMR) from genomic or clinical data. Built for researchers, healthcare pros, and bioinformatics experts, it helps identify resistant microbial strains quickly and accurately using trained ML models. In a world where AMR is a growing threat, we’re proud to have built something with real-world impact.
Project: QueryLens
A full-fledged Data Visualization Agent that could intelligently generate SQL queries across multiple tables and instantly visualize the results.Designed a custom "SQLite server" to simulate real-world multi-table scenarios and manage seamless communication between the frontend and backend. On top of that, we used "LangChain" to power natural language interaction allowing users to generate, understand, and visualize SQL results using plain English prompts.
LeetCode Contributions
Leetcode ・ 385 Question Solved
GitHub Contributions
Research Publications
Under Review – Targeting IEEE / ACM Conference Submission
Author: Rishabh Sharma
Abstract
Federated Learning (FL) enables collaborative model training across distributed clients without sharing raw data, making it a promising paradigm for privacy-sensitive domains such as healthcare, finance, and IoT networks. However, traditional federated learning frameworks assume honest client participation and are vulnerable to model poisoning, unreliable updates, and adversarial manipulation.
Additionally, centralized aggregation servers introduce trust and transparency limitations.This work proposes a Trust-Weighted Federated Learning framework integrated with a blockchain-based audit and reputation mechanism to enhance robustness and accountability in distributed AI training. The system assigns dynamic trust scores to participating clients based on historical model update consistency, gradient similarity metrics, and validation performance. These trust scores are stored on a tamper-resistant blockchain ledger, enabling transparent and verifiable participation records.
Tech Stack
I'm a generalist at heart who can build with anything, but here's the core stack I've spent the most time with:
Recommendations by Clients
I've had the privilege to work with Rishabh, he was an outstanding contributor who brought both technical excellence and genuine intellectual curiosity to every project. He quickly mastered GPU optimization, RAG architectures, and domain-specific NLP to build an AI nursing engine that reduced clinical workflow time by 30%. What set him apart was his ability to navigate ambiguity whether fine-tuning models on medical literature or building compliance automation for enterprise clients, he consistently figured out what needed to be done and executed with minimal oversight. His work ethic, problem-solving mindset, and ability to deliver production-quality systems made him an invaluable team member.
It's rare to find someone who can dive into something as complex as AI-driven trading systems and start contributing meaningfully within weeks. Rishabh came in with almost no background in financial systems but quickly taught himself quantitative modeling, agent architectures, and the nuances of building low-latency trading pipelines.
He doesn't just implement what you tell him he asks the right questions, figures out what actually needs to be built, and delivers production-quality code without constant handholding. If you need someone who can take ownership of ambiguous, technically demanding problems and turn them into real systems, Rishabh has that rare combination of hunger to learn and discipline to execute.
Library
Dev
*and many more, these are just one of my best reads
Thing about me
Beyond engineering and building systems, I find meaning in understanding problems from first principles. Whether it's implementing neural networks without libraries to truly understand backpropagation, my approach is rooted in genuine curiosity about how things work not just how to make them work.
I believe the best engineers are the ones who stay curious beyond their domain. It's the willingness to dive deep into medical, financial, or security infrastructure to understand not just the technical challenge but the human problem that separates products people use from products people depend on.
Pomodoro Timer
You've reached the end! Or have you? Before you vanish into the digital void, I've got a quick Pomodoro Timer to help you focus better on your next big thing (or just to remind you to stop doomscrolling).