I'm

Akash Sharma

11

months

of working experience as a LLM Engineer Intern & AI Engineer Intern

Results-driven AI/ML enthusiast pursuing a B.S. in Data Science at IIT Madras, delivered HFT predictive model, AI agents, automated image captioning, and NeuroVoyager for natural language neuroscience data exploration.

Analytical Problem-Solving

Strategic AI/ML Development

In-Depth Data Science Exploration

View Resume

Companies Worked

2

Innovative AI Engineer Intern at Krsta Consultancy Services and ex-LLM Engineer Intern at IIT Madras, delivering decision-making systems, advanced LLM applications, and innovative workflows for comprehensive data extraction, summarization, and neuro-research.

Projects Completed

5

Developed robust HFT predictive model, AI style-checking agent, and automated image captioning system at Krsta; engineered NeuroVoyager at IIT Madras, enabling efficient natural language-driven neuroscience data querying using LLM-based tools.

Skills & Experience

7 months of experience in a research lab, good knowledge in end to end AI Application Development and predictive Machine Learning models.

My Skills

Python

95%

Javascript

80%

pandas

90%

Linux

80%

AI Agents

90%

RAG/GraphRAG

95%

Docker

85%

FastAPI

90%

AI Engineer Intern

Feb 2025 - Present

Krsta Consultancy Services

LLM Engineer Intern

Jul 2024 - Jan 2025

IIT Madras Brain Research Centre

BS Data Science & Applications

May 2022 - Present

IIT Madras

High School Certificate

2017

SBOA Public School

Secondary School Certificate

2015

Suderashan Public School

Projects

All Projects

NeuroMiner

LLM-based conversational query engine for Human Brain data.

A neuroscience knowledge exploration framework that leverages LLMs, knowledge graphs, and embeddings to connect 22,000+ concepts from 3,248 articles, uncovering hidden patterns and accelerating hypothesis generation.

Tools Used

NLP, RAG/GraphRAG, Vector DB, Neo4j, vLLM, SGLang, AI Agents, FastAPI

View Project

GliaCoder

LLM-powered AI assistant for code generation, completion, debugging, and brain image exploration.

A multi‑platform tool integrated into Code4Humanity as a Chrome extension, web app, and JupyterLab extension, leveraging advanced NLP techniques and the HBP codebase to automate coding tasks and streamline neuro‑imaging workflows.

Tools Used

NLP, Chrome Extension, Web App, JupyterLab, vLLM, AI Agents, FastAPI

Automatic Image Captioning System

LLM-driven captioning for complex scientific images.

A robust pipeline that extracts images and metadata from PDFs, identifies sub-images and contextual references, and uses LLMs to generate accurate short or long descriptions, refined through an automated evaluation system.

Tools Used

Django, PyMuPDF, Gemini API, PostgreSQL, JSON, OpenCV

High-Frequency Trading Predictive Model

LSTM-based model for next-minute stock movement prediction.

Trained on OHLCV data from the top 100 S&P 500 momentum stocks, this model uses 60-minute historical windows with engineered technical indicators and lag features to classify short-term trends as bullish, bearish, or neutral.

Tools Used

Python, LSTM, pandas-ta, scikit-learn, Matplotlib, yfinance, FastAPI, Polygon.io

NeuroVoyager

LLM-based applications—including a Conversational Query Engine(NeuroMiner), NL2SQL system, and AI agent—for neuroscience data exploration and analysis.

Developed NeuroMiner, a Conversational Query Engine, along with an NL2SQL translator and AI agents using Codestral-22B, Nv-embed-v2, and Llama-3.3-70B-Instruct to enable iterative self‑querying, chain‑of‑thought reasoning, and seamless multi‑database operations for streamlined querying, graph exploration, and statistical analysis within the Human Brain Project.

Tools Used

NLP, RAG/GraphRAG, Vector DB, Neo4j, vLLM, SGLang, AI Agents, FastAPI, MySQL, PostGIS

View Project

Bank Telemarketing Campaign Success Prediction

Supervised ML for customer response prediction.

Applied LightGBM, XGBoost, and Gradient Boosting on a Kaggle bank telemarketing dataset with imputation, one‑hot encoding, and standard scaling; evaluated via macro F1‑score to drive balanced targeting and segmentation insights.

Tools Used

Python, scikit‑learn, LightGBM, XGBoost, pandas, NumPy, Jupyter Notebook

View Project

Business Model Analysis of Jockey Outlet

Data-driven retail optimization.

Conducted comprehensive analysis of SR Associates’ Jockey outlet in Guwahati by examining sales data and customer behavior, delivering solutions to optimize inventory management, reduce stockouts, and boost foot traffic.

Tools Used

Data Analysis, Business Data Management, pandas, Matplotlib, Seaborn, Microsoft Excel, Jupyter Notebook

View Project

GliaCoder

LLM-powered AI assistant for code generation, completion, debugging, and brain image exploration.

Tools Used

NLP, Chrome Extension, Web App, JupyterLab, vLLM, AI Agents, FastAPI

Jiggly: Music App

Multi-user music streaming web application.

Built with Vue.js frontend and Flask backend, Jiggly offers daily listening reminders, Redis & Celery caching, and JWT‑based authentication for secure, scalable music streaming.

Tools Used

Python, Flask, SQLite, Vue.js, Redis, Celery, JWT, REST APIs

View Project

Automatic Image Captioning System

LLM-driven captioning for complex scientific images.

Tools Used

Django, PyMuPDF, Gemini API, PostgreSQL, JSON, OpenCV

High-Frequency Trading Predictive Model

LSTM-based model for next-minute stock movement prediction.

Tools Used

Python, LSTM, pandas-ta, scikit-learn, Matplotlib, yfinance, FastAPI, Polygon.io

AI-Enabled Content Management System

LLM-powered editorial assistant for written content.

Built an AI agent to assist with automated style checking and quality assurance in written content, designed for MS Word integration. Style rules were scraped from the Chicago Manual of Style, embedded, and stored in a vector database. The agent evaluates user input and provides character-level corrections aligned with editorial standards.