Hey there, I'm Yash.

Designing & delivering enterprise AI agents in production

I'm an AI Engineer with 2.5+ years designing and delivering enterprise AI agents and workflow automation in sensitive enterprise environments. I'm currently interning at the World Trade Organisation in Geneva, alongside an MSc in Artificial Intelligence at USI Lugano. My work spans RAG pipelines, agentic systems, and responsible AI governance.

Previously at Perimattic, I built conversational AI agents serving 1M+ users and enterprise knowledge-base tools adopted across client operations. You can also find me on Medium and Kaggle.

Download CV Email me cv.pdf · last updated 2026-08-10

projects

Projects & publications

A few things I've built and shipped.

Agentic observability & SRE MCP platform

An MCP-based observability service orchestrating AI agents on n8n for automated incident analysis from Sentry and Prometheus logs — production-grade reliability and debuggability for agent-based systems.

MCP n8n Observability

Private · internal deployment

LLM fine-tuning with QLoRA

Fine-tuned Gemma 7B using QLoRA, then deployed an optimized inference engine with TensorRT-LLM, serving over vLLM.

QLoRA TensorRT-LLM vLLM

Private · internal deployment

RAG application to generate SQL queries

An end-to-end RAG pipeline that turns a natural-language question into a SQL query — it retrieves the schema that best fits the question, then generates the query against it.

RAG SQL generation

Code Live demo

Breast cancer detection (Wisconsin)

End-to-end ML pipeline classifying tumors as benign or malignant, using statistical testing, outlier detection, feature selection, and model optimization.

95.6% accuracy perfect precision

Code View project

NLP Tasks Hub

A multi-task language intelligence playground — a Gradio app showcasing sentiment analysis, toxic comment detection, summarization, question answering, and story generation, built with Hugging Face Transformers.

NLP Gradio HF Spaces

Code Live demo

World Happiness Report 2023 analysis

Analyzed global happiness trends from the 2023 report — data preprocessing, exploratory analysis, feature engineering, and Random Forest modeling with hyperparameter tuning.

EDA Random Forest

Code

experience

Experience

Mar 2026 – present

Intern

World Trade Organisation · Geneva, CH

Designed an AI classification tool using hybrid RAG (BM25 + semantic search, RRF fusion), achieving 73.6% top-1 accuracy across 10,000+ regulatory codes at <$0.01/query, with structured outputs and confidence-threshold routing for reliability.
Built an agentic trade-policy surveillance system using MCP, Google ADK and LLM APIs — automating a fully manual cross-jurisdictional monitoring workflow across 7+ government sources and 6 countries, served with FastAPI.
Benchmarked on-prem deployments of Gemma 3 and Qwen 3.5 models on T4 GPUs to document metrics (tokens/sec, TTFT, TTSL, request latency, and req/sec).

Jan 2023 – Aug 2024

AI Engineer

Perimattic Inc. · Bengaluru

Designed and deployed conversational agentic AI systems with a LangChain pipeline handling 1M+ airport travellers, integrating with legacy backend and cloud-native deployment at 99.95% uptime.
Built enterprise AI agents on Google AI Studio grounded in Docs and SharePoint document corpora, enabling non-technical business users to query internal knowledge bases through natural language — driving measurable adoption across client operations.
Contributed to the AI evaluation framework by integrating DeepEval into LLM-based solutions and regression testing in ML deployments, and built automatic feedback loops by observing production logs.
Built and deployed per-facility demand forecasting models (Facebook Prophet) for airport parking utilization across multiple facilities for a UK parking operator — enabling a dynamic pricing strategy that contributed to 15% revenue growth. Implemented automated drift monitoring to detect degradation and trigger selective retraining, with MLflow for model versioning.
Supported end-user training and adoption: documented workflows, facilitated handoff sessions, and upskilled business teams to operate and iterate on deployed AI agents independently.

Mar 2022 – Jul 2023

Junior AI Engineer

Freelance · Jaipur

Architected a custom RAG solution (Qdrant vector DB) with containerised LLM inference via Docker — 95% hallucination reduction with enterprise-only docs, 99.95% uptime, full GDPR compliance.
Built a real-time computer vision system (YOLO) for automated vehicle number plate detection at an airport — reducing processing time from ~60s (manual lookup) to <5s (12× improvement).

education