ML · AI · DS · Game Development & Design

From real-time shader pipelines to agentic graph-based reasoning systems. Currently pursuing M.Tech in Data Science & AI.

Agentic GraphRAG for Root Cause Analysis  ·  In Progress
Graph-structured retrieval and skill-graph reasoning over cascading failure traces in large-scale cloud microservices. Built on LangChain and LangGraph. Targeting autonomous, explainable RCA without human-in-the-loop triage.

Screenplay GPT-2 Fine-Tuning Study
Comparative study of full-parameter fine-tuning vs. LoRA adaptation for structured creative text generation using GPT-2 Small (124M). Trained across AMD DirectML, Apple Silicon, and NVIDIA T4 (CUDA) backends. Includes a stateful checkpoint recovery from a mid-run hardware preemption at step 5,600/9,272 — resumed with zero loss discontinuity. Both fine-tuned model weights published to Hugging Face Hub. Final validation loss: 1.3194.

StudyBuddy — Local RAG Assistant
Fully offline retrieval-augmented generation over personal document corpora (PDF/PPT). SentenceTransformers for embeddings, FAISS for vector search, Llama 8B via LM Studio for inference. Zero cloud dependency; citation-grounded answers with source chunk attribution.

Multi-Class News Classification
End-to-end NLP pipeline classifying articles across 42 categories (LinearSVC) and 13 consolidated super-categories (LinearSVC + XGBoost). Modular 4-stage architecture: scraping → preprocessing → training → analytics. Ships with an interactive Streamlit dashboard supporting live URL scraping, batch CSV processing, and per-category confusion matrix visualization.

Live Win Probability Engine
Real-time Bundesliga match prediction via Monte Carlo simulation over Poisson-distributed Elo ratings. FastAPI backend with WebSocket delivery of mid-match probability updates. MySQL persistence, Redis caching with mock fallback, and full SQLAlchemy ORM layer.

Languages        C · C++ · Python 
Graphics         OpenGL · GLSL Shader Development · Pixel Art Pipeline
AI / ML          LangChain · LangGraph · FAISS · HuggingFace Transformers
                 Fine-tuning (Full + LoRA/PEFT) · RAG · Graph Neural Reasoning
ML Ops           Checkpoint Recovery · Mixed Precision (FP16) · Multi-backend Training
Backend/DB       FastAPI · WebSockets · SQL · Redis 

Two fine-tuned models and tokenized dataset used to fine tune the model on Hugging Face — GPT-2 Small adapted for screenplay generation, trained across multiple hardware backends with full training telemetry documented.
→ huggingface.co/raghavnimbalkar1

Designing procedural shaders and pixel art pipelines for a retro-aesthetic game project  ·  Researching agentic graph traversal for fault localization in distributed systems  ·  Learning OpenGL internals at the pipeline level

LinkedIn  ·  Hugging Face  ·  Email  ·  Kaggle