Akash Varma Datla
AI Developer with a passion for building intelligent systems using Python, FastAPI, and machine learning frameworks.
AI Developer with a passion for building intelligent systems using Python, FastAPI, and machine learning frameworks.
I'm Akash Varma Datla, an AI Developer with a strong foundation in applied machine learning, full-stack AI systems, and intelligent agent workflows. I completed my B.Tech in Artificial Intelligence from SRM Institute of Science and Technology, where I honed my skills across the AI spectrum—from core algorithms to real-world deployments.
My portfolio includes building multi-agent conversational systems like PitLane Insiders for F1 analytics, end-to-end PDF Q&A platforms like PDFSense, and deploying intelligent pipelines using LangChain, HuggingFace, and FastAPI. I’ve also worked on real-time traffic detection with YOLOv8 during my research internship at the Taiwan–India Big Data Lab.
With skills spanning PyTorch, TensorFlow, LangChain, FastAPI, Docker, AWS, and GitHub Actions, I focus on transforming AI ideas into usable, deployed products. I love solving problems where AI meets interaction — chatbots, smart APIs, and creative tools.
đź“Ť Mumbai, Maharashtra, India
LangChain
AgnoI built PitLane Insiders, a multi-agent AI system that uses OpenAI LLMs, Agno tools, and a custom MCP server to answer natural language questions about live Formula One data. It integrates telemetry, pit stop strategy, lap insights, and weather using FastF1, Polars, and FastAPI. The system also features a Discord bot for conversational access.
I developed PDFSense, an AI-powered PDF Q&A system using LangChain, ChatGroq, FAISS, and Hugging Face Transformers. It features a custom RAG pipeline for semantic search and context-aware answers, deployed with Streamlit on Hugging Face Spaces and automated via GitHub Actions.
I built Monetara, a multi-agent AI system using the Agno SDK for financial data retrieval and web search. The agents collaborate to handle specific tasks and are deployed as a locally hosted Playground app via Agno Cloud.
I developed a predictive model for customer churn using an Artificial Neural Network built with TensorFlow. By analyzing customer profiles and interaction data, the model classifies churn likelihood. The project demonstrates model training, evaluation, and deployment using Flask for web integration.
I built a Flask-based machine learning app that predicts used car resale prices using inputs like price, mileage, fuel type, and vehicle age. The model, trained with scikit-learn, achieved 96.12% accuracy and serves real-time predictions through a user-friendly web interface.
I developed a lyrics-based sentiment analysis system using LSTM and BERT models to classify emotions across diverse music genres and languages. By preprocessing multilingual text with tokenization and TF-IDF, and fine-tuning for accuracy, the project delivers a modular framework with potential applications in music recommendation, emotional wellness, and creative tools.
Student research intern
Authors: U. Sakthi, Aman Parasher, Akash Varma Datla
Conference: 2025 4th International Conference on Deep Sciences for Computing and Communications (IconDeepCom 2025, Hybrid Mode)
Developed a real-time underwater monitoring system that detects and tracks multiple fish species using YOLOv12, ByteTrack, and BoT-SORT. A user-calibrated pixel-per-meter module enables accurate speed estimation, improving ecological behavior analysis. The model achieved 89% accuracy and was deployed via a Streamlit app for marine researchers. This work advances automation in marine biology and fisheries research.
Authors: U. Sakthi, Aman Parasher, Akash Varma Datla
Conference: 2025 9th International Conference on Smart Trends in Computing and Communication (SMARTCOM 2025, Hybrid Mode)
Built a fine-grained ship classification system trained on the FGSC-23 dataset, using ResNet50v2, EfficientNet, and MobileNetV2. The model accurately classified six key ship categories for both civilian and defense applications. ResNet50v2 achieved 87.9% accuracy and 90.5% precision, proving its real-world viability for maritime surveillance. The project emphasizes practical deployment for security, rescue, and naval monitoring.
