AI / GenAI Engineer

Retrieve first,
answer second.

Final-year CS undergraduate who builds LLM agents, retrieval pipelines, and computer-vision systems meant to hold up in production — from enterprise copilots to a weapon-detection model running at 99% accuracy.

99%weapon detection accuracy
98%violence detection accuracy
3production-oriented AI systems shipped
ask-about-ashish.rag
self-rag · grounded

This box works the same way my projects do: it retrieves relevant context first, then generates an answer grounded in it. Try a question above.


selected systems

What I've shipped

Three systems spanning agentic retrieval, multimodal search, and real-time vision — each built to run, not just to demo.

Multi-tool · Async

Enterprise AI Knowledge Copilot

  • LangGraph agent orchestrating MCP tools for ticketing, employee lookup, and policy retrieval.
  • Self-RAG/CRAG verification loop grounds answers in retrieved context to cut hallucinations.
  • Async multi-tool backend (FastMCP + PostgreSQL/pgvector) for concurrent enterprise workflows.
LangGraphMCPSelf-RAGPostgreSQLFastAPILangSmith
Vision + Language

Multimodal RAG Product Assistant

  • Vision-language models paired with semantic retrieval for natural-language product search.
  • Automated image-to-text pipeline generates searchable metadata, removing manual cataloging.
  • FAISS-backed vector index for fast, relevant retrieval over product embeddings.
LangChainFAISSVLMsStreamlit
99% accuracy

CCTV Weapon & Violence Detection

  • YOLOv8 tuned via targeted augmentation and hyperparameter search on custom CCTV footage.
  • 99% weapon-detection accuracy; a fine-tuned RS3D spatiotemporal model adds 98% violence detection.
  • Full pipeline: frame preprocessing → inference → alerting, built for real surveillance feeds.
YOLOv8OpenCVPyTorchRS3DFFmpeg

experience

Where I've worked

Apr 2026 — Jun 2026
Remote

Machine Learning Intern — Team Lead

Intakeoff
  • Trained and optimized real-time computer-vision workflows using OpenCV and YOLOv8 for object detection and image classification on custom datasets.
  • Engineered an end-to-end inference architecture — preprocessing, inference, and post-processing — for production deployment.

stack

Tools I reach for

AI & LLM systems

LangChainLangGraphLangSmithLlamaIndexMCPSelf-RAG / CRAGAgentic AIPrompt Engineering

Computer vision

YOLOv8OpenCVPaddleOCRFFmpeg

Machine learning

PyTorchScikit-learnXGBoostPandasNumPy

Deployment & languages

PythonC / C++JavaScriptSQL / PostgreSQLFastAPIFlaskStreamlitGit

education

Background

ABVGIET Shimla, Himachal Pradesh
B.Tech, Computer Science · Jun 2023 – Present
GPA 7.97
Snower Valley Public School, Kullu
Senior Secondary Education · Mar 2021 – Mar 2023
94%