AI/ML Engineer (LLM Apps + MLOps) | Python | FastAPI | Airflow | Vector Search

I build production-focused AI systems: LLM agents, RAG + hybrid search, multilingual data pipelines, and deployable ML services with strong attention to latency, cost, and reliability.

• 🧠 LLM Apps: LangGraph / LangChain, RAG, tracing/observability
• ⚙️ Backend: FastAPI, async, WebSockets, worker pools
• 🗃️ Search/Data: Vector DBs (e.g., Weaviate/Chroma), SQL/NoSQL, pipelines (Airflow)
• 🚀 MLOps: Docker, CI/CD (GitHub Actions), monitoring (Prometheus), AWS

• Built a multilingual document-filtering pipeline (batching + caching + async) to replace a slower LLM-heavy workflow.
• Reduced report generation runtime by optimizing pipeline stages and cutting unnecessary external API calls.
• Worked on hybrid search features (vector + keyword) and implemented boolean/phrase search for diverse unstructured sources.
• Built LLM agent workflows and automated report generation (PDF/DOC), with tracing for debugging and performance.

• Fine-tuned open-source LLMs for domain tasks using PEFT (LoRA/QLoRA) and built training/eval scripts.
• Improved inference latency using post-training quantization and validated quality vs baseline.
• Deployed containerized inference services (FastAPI + Docker) and integrated CI/CD with GitHub Actions.

A medical documentation assistant prototype:

• Speech-to-text (Whisper) + structured note generation
• DICOM/X-ray report generation (prototype workflow)
• FastAPI backend + React frontend + Neo4j graph relationships
  Stack: FastAPI, React, Neo4j, Docker, Whisper

Note: This is a personal prototype. I’m happy to share architecture details, benchmarks, and demos during interviews.

End-to-end pipeline to detect malicious URLs:

• Airflow DAGs for pipeline orchestration
• MLflow experiment tracking + model versioning
• Dockerized service + AWS deployment
  Stack: Random Forest / XGBoost, MongoDB, Airflow, MLflow, Docker, AWS

RAG-based pricing analysis over a large product catalog:

• Vector search with sentence embeddings + ChromaDB
• Agentic analysis workflow
  Stack: RAG, Vector DB, Sentence Transformers, LangChain

Explored a transformer attention variant that injects a learnable locality bias to reduce early training noise and improve stability.

Languages: Python, SQL, JavaScript
ML/DL: PyTorch, TensorFlow, Scikit-learn, Hugging Face, XGBoost
LLM Systems: LangGraph, LangChain, RAG, evaluation + tracing
Data/MLOps: Airflow, Docker, GitHub Actions, MLflow, Prometheus
Databases: PostgreSQL, MongoDB, Neo4j, ChromaDB (and other vector DBs)

• Microsoft Certified: Azure AI Engineer Associate (AI-102)

• Email: amanagnihotri902@gmail.com
• LinkedIn: https://www.linkedin.com/in/aman-agnihotri004/
• GitHub: https://github.com/its-amann

• Production-grade RAG evaluation + retrieval testing
• Faster inference + caching strategies for LLM apps
• Better data quality + pipeline reliability