Headline

Applied AI Engineer | Multimodal Parsing Pipelines, Retrieval Systems, Multi-Agent Workflows, Durable LLM Systems

About

I work on AI systems built around parsing, retrieval, orchestration, and structured execution. My work has included multilingual categorization, product deduplication, hybrid search, and batch ML systems using embeddings, Elasticsearch, BM25, and API-driven workflows. More recently, I’ve been building systems that process large PDFs and specification-heavy documents into structured JSON, CSV, and markdown, then use that data in multi-step LLM workflows rather than relying on raw document chat. That includes multi-agent orchestration, persistent memory layers, fallback paths between structured outputs and source documents, prompt-safe workflow design, LLM-as-judge style evaluation loops, and traceable reporting. I’m most interested in systems where model quality depends as much on data flow, memory, retrieval, and execution design as it does on the model itself.

Location

Gurgaon

AI stack

cursor

hugging face

langchain

langgraph

llmapi

ollama

Work experience

Applied AI Engineer at Paladio AI, Sep 2023 - Present · 3 yrs 1 mo, Bengaluru, Working on Paladio AI product and research initiatives. Multimodal systems | Structured parsing | Hybrid retrieval | Multi-agent systems | Memory and fallback design | LLM evaluation

AI Engineer 1 at IQVIA, May 2023 - Aug 2023 · 4 mos, Bengaluru

Machine Learning Engineer at FITTR, Nov 2021 - Apr 2023 · 1 yr 6 mos, Pune Division, Developed a prototype for an Exercise Tracker and Pose Detection System on iOS and Android, featuring real-time repetition tracking and posture correction. Utilized advanced algorithms to accurately count exercises and detect user posture, enhancing safety and effectiveness. The prototype demonstrates cross-platform compatibility and provides instant feedback to help optimize form and minimize injury risks. Integrated the Elastic Search with the system for diet tool functionality Created a Hybrid recommendation system for Coach recommendation based on the user interaction. Multiple factors were taken into consideration and their priorities were set in order to prioritise relevant elements. The project is based on automated pipeline so that after a certain interval it reconfigure itself to ever changing user interactions.

Machine Learning Engineer at Socialkyte, Feb 2021 - Nov 2021 · 10 mos, Noida, Uttar Pradesh, India, Profile Classification: Architected a layered, multi-label classification system utilizing Transformer models, Zero-Shot Learning, Word2Vec, and LDA, refined with Universal Sentence Encoder and HDBSCAN for superior data categorization and processing. Recommendation System: Assembled a recommendation engine that merges textual data through advanced NLP frameworks, boosting accuracy in profile connections.

Education

Bachelor of Technology, Computer and Communication Engineering, The LNM Institute of Information Technology