name: "Harshprabha"
role: "AI/ML Engineer · AI Safety Researcher · Full-Stack Engineer"
based_in: "New Delhi, India"
education: "B.Tech ECE (AI Specialisation), IGDTUW — 2025–2029"
focus:
  - Evaluation infrastructure for LLM safety & alignment
  - Chain-of-thought faithfulness & agentic compliance research
  - Full-stack, AI-integrated product engineering
  - Research communication & technical leadership

I'm an engineering student and researcher building at the intersection of LLM evaluation, AI safety, and applied full-stack engineering. My work centers on making model behavior legible — measuring the gap between what models say they'll do and what they actually do, and probing whether reasoning traces reflect genuine awareness rather than post-hoc justification.

Alongside research, I ship production-shaped systems: agentic evaluation harnesses, real-time financial intelligence pipelines, and civic-tech platforms — with an engineering mindset that treats reproducibility, testing, and clean architecture as non-negotiable.

Open To: AI Safety Research Roles · Research Internships · Technical Collaborations · Full-Stack / AI Engineering Roles

Generative AI Research Intern · MacroedTech May 2026 – Present · Remote

Investigating LLM evaluation methodologies, foundation model capabilities, and agentic AI failure modes across GenAI deployment contexts.

• Synthesized findings into domain-specific technical reports on real-world GenAI deployment risk
• Conducted systematic literature review across prompting strategies, RAG pipelines, and multi-step reasoning
• Distilled technical insights into structured analyses bridging engineering depth and stakeholder communication

LLM Evaluation RAG Technical Writing Risk Analysis

Research Intern — AI Safety & Alignment · ACM May 2026 – Present · Remote

Contributing to LLM behavioral evaluation research probing reasoning faithfulness and alignment-relevant failure patterns.

• Designed experiments probing reasoning faithfulness in model scratchpad outputs
• Documented alignment-relevant failure patterns across evaluation runs
• Synthesized literature on model compliance and behavioral consistency

AI Alignment Experimental Design Reasoning Faithfulness

President · Youth Economy Lab (YEL), IGDTUW Chapter Apr 2026 – Present

• Established the IGDTUW chapter from the ground up
• Spearheaded workshops on AI governance, policy, and the economic implications of frontier AI

Leadership AI Governance Community Building

Research Coordinator · E-Cell IGDTUW Nov 2025 – Present

• Authored analytical newsletters on emerging tech trends and startup ecosystems
• Built research communication skills across technical and non-technical audiences

Research Communication Startup Ecosystems

Tech Coordinator · IEEE IGDTUW Nov 2025 – Present

• Led Open Source Week with 100+ participants
• Organized technical sessions on version control and open-source contribution workflows

Developer Advocacy Event Engineering Open Source

_{Live GitHub widgets — if a card shows blank, refresh the page; the underlying service occasionally needs a moment to render.}

learning:
  - Mechanistic interpretability & scratchpad faithfulness auditing
  - Advanced experimental design for multi-turn LLM oversight studies
  - Distributed systems observability for multi-agent architectures

building:
  - Refusal Recovery Under Mid-Conversation Oversight Injection (AI safety research)
  - SigNoz-integrated observability tool for CoT faithfulness & cascading agent failures
  - Stated vs. Actual Compliance framework (agentic evaluation harness)

exploring:
  - Alignment faking in small language models
  - Cross-model transferability of interpretability signals
  - AI governance frameworks for the South Asian context

open_to:
  - AI Safety Research Internships & Fellowships
  - Graduate research collaborations
  - Full-stack / AI engineering roles

"Evaluation infrastructure is the prerequisite, not the afterthought, of safe AI deployment."