Credentials & The Archives

A record of continuous learning and applied research. Fusing cloud-scale infrastructure with neural network inquiry.

Holographic Certifications

AWS Certified Cloud Practitioner

Foundational understanding of AWS Cloud services, architecture, pricing, and support. Validated cloud literacy and deployment best practices.

ID: AWS-CCP-2024

Database Management System (DBMS)

National Programme on Technology Enhanced Learning (NPTEL) certification from Indian Institute of Technology, Kharagpur. Comprehensive coverage of DBMS concepts, SQL, normalization, and transaction management.

ID: NPTEL22CS51S43582362

Introduction to C++ Completion Certificate

Coding Ninjas certification covering C++ fundamentals, object-oriented programming, data structures, and algorithmic problem-solving.

ID: 1462393dd86c5323cfd272856817319bb0c2c89

Digital Archive: Publications

Index 2023 – 2024
Research PaperJun 2024

Evaluating the Performance of Fake Face Detection Methods in Forensics

Developed an ML model using Local Binary Patterns and classification techniques, compared multiple algorithms for fake face detection in forensic applications.

IJITCE

Read Manuscript
Research PaperMay 2023

Optimizing Prediction of Security Data Using Feature Selection and Ensembling

Built a machine learning-based network security model to detect intrusions with improved accuracy using ensemble machine learning algorithms on large datasets.

IJITCE

Read Manuscript
// Lab Notebook

Active Research

Investigating the boundaries of AI reliability, reasoning, and security.

completedSep 2025 – Dec 2025

Hybrid Multi-Retriever Orchestration for Long-Term Conversational Memory

Built a RAG orchestration system routing queries across 5 retrievers (BM25, TF-IDF, FAISS, SVM, Time-Weighted) using intent classification and rule-based decision logic. Improved Recall@5 by +18 pts (0.65 to 0.83) on LongMemEval with Qwen2.5-7B and +4 pts on Locomo with LLaMA-3.1-8B.

RAGMemoryRetrievalQwen2.5-7B
ActiveJan 2026 – Present

Agentic Workflow Orchestration for Multi-Step Task Automation

Designing and building agentic AI systems that decompose complex tasks into executable sub-goals, leveraging tool-use, planning, and memory modules. Exploring architectures for autonomous agents that can reason, plan, and act across multi-step workflows with minimal human intervention.

AgentsLLMOrchestrationTool-Use
completedAug 2025 – Oct 2025

STaR: Self-Taught Reasoner for Mathematical Reasoning

Fine-tuned Llama-3.2-3B-Instruct using STaR rationale-bootstrapping on NVIDIA A100 GPUs, achieving 46.2% on GSM8K (+65% relative gain over zero-shot CoT). Generated 3.5K synthetic hint-conditioned training samples via self-improvement loops, outperforming vanilla SFT (39.2%) by 7 absolute points.

ReasoningFine-tuningLlama-3.2GSM8K

Hackathons & Wins

Combat-tested engineering in high-stakes competitive environments.

JPMorgan Code for Good

National / Remote

1ST PLACE
RASAPythonNLP

NASA Space Apps Hackathon

Chandigarh, India

FINALIST
Space TechData VizPython

Model United Nations

Hyderabad, India

1ST PLACE
DebatePolicyResearch