Applied AI/ML Engineer with end-to-end ownership of RAG pipelines, LLM agents, and production ML
systems. Built a semantic search engine over 42,549 papers benchmarked at 2,979 QPS (p50 = 0.31 ms, p95 =
0.53 ms) and a medical NLP/RAG system deployed on GCP. Hands-on across the full stack: data ingestion →
vector retrieval → LLM serving → production deployment. Stack: FastAPI · FAISS · LangChain · Docker · GCP · MLflow ·
HuggingFace.
Technical Skills
AI Agents & RAG:
LangChain, LangGraph, LlamaIndex, ReAct, Multi-Agent, RAGAS
LLMs & GenAI:
HuggingFace, BERT/GPT/LLaMA, Fine-Tuning (LoRA/QLoRA), Groq API
Vector & Search:
FAISS (IVFFlat/HNSW/PQ), ChromaDB, Weaviate, Hybrid Search, Reranking
ML & DL:
PyTorch, TensorFlow, Scikit-learn, CNNs, RNNs, Ensemble, XGBoost
Cloud & MLOps:
GCP (Vertex AI, GCS), AWS (EC2, S3), Docker, MLflow, DVC, GitHub Actions
Data & Backend:
FastAPI, PySpark, Pandas, SQL (Advanced), PostgreSQL, Redis
Experience
- Designed and shipped Well.AI end-to-end: raw patient data ingestion → HuggingFace embeddings →
ChromaDB → LLM clinical QA, served via FastAPI on GCP with Docker Compose and Supabase session
logging.
- Engineered prompt workflows for clinical health reasoning; structured retrieval context windows
to improve response faithfulness and reduce hallucination rate in medical QA tasks.
- Translated health business requirements into ML solutions in collaboration with domain stakeholders; delivered
across dev, staging, and prod with governance-ready deployment.
- Built automated ML pipeline on a high-dimensional tabular dataset (50K+ SNP features): feature
engineering → ensemble training (XGBoost + Random Forest) → hyperparameter tuning, achieving 87.6%
classification accuracy.
- Deployed interpretable Streamlit dashboard for non-technical stakeholders; implemented
DVC for full data versioning and experiment tracking on GCP.
Projects
sentence-transformers · FAISS IVFFlat · FastAPI · Groq LLaMA-3 · ChromaDB · Docker · Streamlit
- Built full retrieval pipeline over 42,549 arXiv papers (20 domains): ingestion → 384-dim
embeddings → FAISS IVFFlat (62.3 MB index) → RAG summarisation; benchmarked at p50 = 0.31 ms, p95 = 0.53
ms, 2,979 QPS, production-grade throughput on a low-resource deployment.
- FastAPI backend with LRU embedding cache cut repeated query latency from 1,100 ms →
0.94 ms (55× speedup); integrated Groq LLaMA-3.3-70b agentic pipeline generating structured
4-part paper summaries in ~1,200 ms.
FastAPI · ChromaDB · Groq Llama-4 Scout · SentenceTransformers · LangChain · Streamlit
- Production RAG pipeline over 70 MB+ enterprise ticket logs: PDF parsing → chunking →
SentenceTransformer embeddings → LangChain ReAct agent with tool use, achieving p95 retrieval under 200
ms.
- Implemented agent memory, planning, and multi-step execution loops; enabled natural-language
querying over structured enterprise data — applicable to support automation and operational analytics.
TensorFlow · CNN · Transfer Learning · Data Augmentation · Docker
- Automated preprocessing → augmentation → CNN training → evaluation pipeline; achieved 93%
accuracy on 4-class MRI brain tumor classification with transfer learning; containerised inference via
Docker.
Education
Master of Computer Applications (MCA)
University of Madras, Chennai · 2023–2025 · GPA 7.38/10
Specialisation: Machine Learning, Deep Learning, Data Science
B.Sc. Biotechnology
RKM Vivekananda College, Chennai · 2020–2023 · GPA 7.2/10
Foundation in statistics, computational methods, and biological data analysis
Achievements, Leadership & Certifications
- Mentor & AI Educator — Guided HSC students (3 yrs 70+ students) on AI fundamentals and career
paths
- Vice President, Rotaract Club of RKMVC — led 5+ AI awareness events across Chennai
- 2nd Place — AI Front-End Hackathon 2025, MCC Chennai (25+ competing colleges)
- Published technical articles on EDA, NLP pipelines, and ML on Medium
- AWS Data Engineering Bootcamp (Udemy, 2024) · Deep Learning Specialization (Coursera, In
Progress)