I'm a first-year engineering student at IGDTUW, building intelligent systems at the intersection of NLP, LLM orchestration, financial data, and AI policy. I like turning messy, real-world data into something that actually makes decisions better — and thinking critically about the governance implications of the systems I build.

• 🔬 Researching agentic prompting strategies for financial sentiment — achieved +27% accuracy gain with Chain-of-Thought
• 🧠 Working with FinBERT, Groq (LLaMA 3.1), Claude API, and volatility-aware feature engineering
• 🌱 Deepening expertise in LLM orchestration, agentic workflows, and cloud architecture
• 🏆 National finalist — Top 10 of 581 teams at HackwithMAIT 6.0 (Top 1.7%)
• 📡 Open source contributor — GirlScript Summer of Code (GSSoC) & Social Summer of Code (SSoC) 2025
• 📬 Reach me at harshprabha95@gmail.com

An agentic evaluation framework benchmarking LLM prompting strategies on financial text.

• Benchmarked 5 prompting strategies (Zero-shot, Few-shot, CoT, Anchor, Agentic) on 300 labeled financial sentences
• Improved sentiment classification accuracy from 62% → 89.67% using Chain-of-Thought prompting (+27% gain)
• Identified neutral sentiment bias; refined approach with anchor prompting; automated a fully reproducible pipeline

An end-to-end NLP pipeline for financial news sentiment classification using FinBERT, deployed on Streamlit.

• Designed a Sentiment Reliability Index (SRI) combining source credibility, recency, and model confidence (score: 67.9)
• Engineered a feature pipeline integrating news sentiment, VIX volatility data, and price trends
• Automated real-time data ingestion via REST APIs; preprocessing with Pandas & NumPy; inference in PyTorch

A responsive, accessible community learning web app — built in 24 hours at HackwithMAIT 6.0.

• Top 10 nationally out of 581 teams (Top 1.7%)
• Built with HTML5, CSS3, JavaScript + Firebase (Auth & Firestore real-time database)

Languages        Python · Java · JavaScript
AI / ML          PyTorch · FinBERT · HuggingFace · Pandas · NumPy · Sentiment Analysis · Feature Engineering
LLM & Agentic    Prompt Engineering · Chain-of-Thought · Agentic Workflows · LLM Orchestration · Groq (LLaMA 3.1) · Claude API
Web              HTML5 · CSS3 · REST APIs · Web Scraping
Database         Firebase (Firestore · Authentication)
Tools            Git · GitHub · VS Code · Google Cloud Platform · Streamlit

• 🌍 Open Source Contributor — GirlScript Summer of Code (GSSoC) & Social Summer of Code (SSoC) 2025 · PRs reviewed and merged by maintainers across AI/ML and web projects
• 🏛️ Chapter Lead @ Youth Economy Lab, IGDTUW (Nov 2025 – Present) — Founded the chapter; organized workshops at the intersection of technology, economics, and governance
• 📝 Research Coordinator @ E-Cell IGDTUW (Nov 2025 – Present) — Researching tech trends and startup ecosystems; authoring weekly analytical newsletters
• 📡 Tech & Event Management Coordinator @ IEEE IGDTUW (Nov 2025 – Present) — Managed IEEE Open Source Week (100+ participants); structured workshops and volunteer teams

Intersection of AI systems and public policy — data protection & privacy, generative AI governance, and AI cybersecurity risk. Focused on how technical realities of LLMs (hallucination, bias, adversarial vulnerability) translate into regulatory frameworks, particularly in the South Asian context.