Looking for a job

Yashraj Jyoti Sanjay Kastode

@yashrajjyotisanjayka

Message

I build reliable AI and NLP applications, from RAG chatbots to scalable inference systems.

India

Message

What I'm looking for

I'm looking to build reliable AI, NLP, and backend systems where I can improve model performance, inference scalability, and real-world product outcomes.

I've built AI products including EduBot, a domain-constrained conversational AI system using LangChain, LLMs, Django, ChromaDB, and Flutter. I improved its query response time by an estimated 35% and sustained 99.95% availability under load through backend and API rate-limit optimization.

I've also deployed Hindi NLP inference with mT5, Django REST Framework, Celery, and Redis, reducing latency by 40%, and built computer-vision and healthcare projects. My work includes a 90% accurate Indian Sign Language classifier, a drug-interaction checker, and a top-132 finish among 80,000+ participants in the Amazon ML Challenge 2025.

Experience

Work history, roles, and key accomplishments

Hindi AI Text Summarizer | Scalable NLP Inference Deployment

Ongoing

Delivered an abstractive Hindi summarization pipeline with Google mT5-Base Transformer, achieving high ROUGE-score fidelity across a 500+ sample dataset. Architected a scalable model inference pipeline using Django REST Framework, enabling real-time NLP predictions and seamless frontend integration.

Meta Trader 5 Transformers Django Rest Celery Redis NLP Model Inference Asynchronous Task Queues (Dramatiq)

EduBot | Domain Constrained Conversational AI

Ongoing

Spearheaded chatbot development using LangChain and LLMs, targeting 3+ educational curricula via semantic search on ChromaDB. Optimized backend performance and API rate-limit strategies on a Django + ChromaDB production system, sustaining 99.95% system availability under load.

Lang Chain LLM Django ChromaDB Flutter Semantic Search API Rate Limiting System Availability

Hasthakshar | Indian Sign Language Recognition

Ongoing

Engineered a proprietary dataset of 1000+ labeled Indian Sign Language samples. Developed a deep learning classification architecture utilizing TensorFlow, achieving a 90% validation accuracy.

Computer Vision Mediapipe OpenCV TensorFlow PyTorch Deep Learning Image Processing

Med-Matrix | Rule-Based Drug Interaction Checker

Ongoing

Designed a rule-based interaction engine backed by 2 structured datasets – drugs.csv and interactions.csv. Implemented a brand-name-to-component resolution layer that maps user-selected medicines to their active salts, generates all C(n,2) component pairs, and cross-references them against 300+ known interaction records to return severity-ranked warnings.

Python Django Rest Combinational Logic Pandas Data Processing