Portrait of Harshit Dabas

Hi there!

I'm Harshit, an AI engineer focused on generative AI systems, retrieval-augmented generation and multimodal document intelligence. I build production-grade workflows that combine vector search, reranking, LLM orchestration and dependable backend delivery.

My work spans applied machine learning, evaluation and product-minded engineering, with a strong interest in turning complex AI pipelines into practical tools that are scalable, measurable and easy to use.

Experience

Professional experience across applied AI, ML and production delivery.

ALKF - Remote Jan 2026 - Jul 2026

Artificial Intelligence Engineer Intern

Built retrieval and document-intelligence systems for production AI workflows with a focus on search quality, structured metadata and multimodal retrieval readiness.
  • Delivered about 94% retrieval accuracy by engineering a RAG pipeline over 255+ documents using semantic chunking, hybrid search and Qwen embeddings.
  • Improved search relevance by 35% with BAAI/BGE cross-encoder reranking, Reciprocal Rank Fusion and top-result narrowing for stronger final answers.
  • Built a document intelligence pipeline that extracted full-page and embedded images with page, section and proximity metadata stored in Supabase.
GC Consulting - Remote Sep 2025 - Mar 2026

Machine Learning Engineer Intern

Worked on model research, benchmarking and optimization experiments across training setups, feature pipelines and deep learning frameworks.
  • Benchmarked optimization approaches including backpropagation, DFA and LocoProp across epochs, batch sizes and learning-rate settings.
  • Reached 90% to 97% predictive accuracy through feature engineering and model tuning in PyTorch, TensorFlow and scikit-learn.
  • Improved performance by 5% to 10% by combining the strongest benchmarked behaviors into a hybrid optimization approach.
gaDgaP AI - Remote Oct 2025 - Dec 2025

Generative AI Engineer Intern

Supported AI education products by building course assets and collaborating on curriculum for practical, high-engagement learning experiences.
  • Created illustrations, flowcharts, Mermaid diagrams and coding exercises for Voice AI and Agentic AI learning content.
  • Contributed to course materials that supported a Udemy learner base of more than 9,000 students.
  • Worked directly with the founder and content team to launch AI-assisted learning resources with clearer curriculum structure.
Proficiency

The tools and disciplines I keep returning to.

Languages

  • Python, SQL and R
  • C, Java, HTML and CSS
  • Production-ready scripting and web fundamentals

AI and ML

  • Machine learning and deep learning
  • NLP, computer vision and evaluation
  • Feature engineering, model tuning and agentic automation

Generative AI

  • LLM applications, RAG and prompt engineering
  • LangChain, vector search, hybrid retrieval and n8n
  • Reranking pipelines, structured output design and ElevenLabs

Frameworks and Tools

  • PyTorch, TensorFlow, Flask and FastAPI
  • Docker, PostgreSQL, MongoDB and Supabase
  • Git, GitHub, Jupyter and production delivery workflows
Projects

Products I've built and continue to refine.

Voice journaling platform Live

Remembrance

An AI voice journaling platform built around voice capture, retrieval and reflective summaries. The system uses a RAG pipeline to generate contextual weekly, monthly and yearly insights while supporting production deployment and scale.
Visit project
Prompt systems In progress

Prompt Library

A curated collection of prompt patterns, evaluation notes and reusable templates for reliable LLM outputs across analysis, generation, extraction and workflow automation.
Open page

Chatbots

Live

AI Avatar

A conversational digital twin designed to answer questions about my background, work, projects, strengths and career direction through a polished portfolio chat experience.
Open chatbot
Live

PC Retail Chatbot

A retail assistant grounded in an INR PC-parts catalog. It estimates complete build costs and recommends compatible PC configurations based on a customer's budget and intended use.
Open chatbot
Live

Law Advisor Chatbot

A legal-information assistant for Indian legal research. It uses current web research and Indian Kanoon legal documents to provide clearly structured answers with sources.
Open chatbot

AI Workflows

Live

Serper n8n Telegram integration

A Telegram AI assistant workflow that uses Serper for current online research before responding to authorized users.
Explore workflow
Academic workflow tooling Live

AI Grading Assistant

A grading workflow that evaluates PDF, DOCX and PPT submissions against rubric criteria and generates structured feedback with multi-model support for scalable academic review.
Visit project
About

I like building thoughtful systems, not just flashy demos.

What keeps me engaged is the gap between a clever AI demo and a dependable user experience. I enjoy the engineering work required to close that gap: better prompts, safer workflows, cleaner interfaces, stronger retrieval and more reliable outputs.

I'm especially interested in products that combine machine learning with strong product design, whether that means document intelligence, educational tooling, assistants or practical knowledge systems.