CV
Download my CV or view it below.
Contact Information
| Name | Md Ayan Arshad |
| Professional Title | AI Engineer |
| ayanarshad2002@gmail.com | |
| Phone | +91 7377869686 |
| Location | Chennai, India |
| GitHub | https://github.com/AyanArshad02 |
Professional Summary
GenAI Engineer with ~2 years of production experience building reliable backend systems for RAG and agentic workflows. Architected a multi-tenant RAG chatbot at Softeon serving 500+ enterprise tenants with hybrid retrieval, tenant isolation, and observability. Built FastAPI-based RAG systems with Claude MCP integrations for agentic tool use. Passionate about context engineering, agent workflows, and shipping high-quality production AI systems. Available for US EST / EU hours.
Experience
-
2025 - Chennai, India
GenAI Engineer
Softeon
- Multi-Tenant RAG Chatbot — 500+ Enterprise Tenants: Architected and led delivery of a production multi-tenant WMS chatbot; reported directly to AVP of Data Science.
- Designed HLD and LLD for multi-tenant architecture using AWS (Cognito, S3, DynamoDB, Lambda, EC2); defined data flows, access boundaries, and deployment strategy.
- Solved tenant isolation at the vector DB layer (Pinecone namespaces), eliminating cross-tenant data-leak risk across 500+ enterprise tenants.
- Built hybrid retrieval with reranking and grounding validation; pushed faithfulness from 0.67 → 0.91 and unsupported claim rate below 4%.
- Led a 2-engineer team: task breakdown, sprint planning, code reviews, and on-time delivery.
- AI-Driven Ticket Resolution System: Led end-to-end design and delivery for Softeon’s WMS support division, owning the project from architecture (HLD/LLD) to production rollout.
- Built a statistically evaluated semantic retrieval system over historical tickets and documentation using PostgreSQL and ChromaDB, significantly reducing manual lookup time and unresolved ticket backlog.
- Performed EDA on historical support data to analyse ticket distributions, resolution patterns, and retrieval effectiveness; optimised relevance thresholds and retrieval depth.
- Designed evaluation workflows to assess retrieval precision, response usefulness, and failure cases, enabling continuous system improvement.
- Implemented monitoring and logging to support reliability and future extensibility.
-
2024 - 2025 Remote
GenAI Engineer
Second Brain Labs (SBL)
- Built a LinkedIn outreach chatbot integrated with the LinkedIn API: personalised messages, multi-turn conversation handling, and lead qualification across multiple client accounts simultaneously.
- System ran on GPT-4 with real campaign traffic.
Awards
-
2024 Topped Programming in Python & System Command
IIT Madras
Topped the Programming in Python & System Command course.
-
2024 Machine Learning Practice Coding Exam
IIT Madras
Scored 85+ in the Machine Learning Practice Coding Exam.
-
2024 ML Project — System Threat Forecaster
IIT Madras
Scored 93/100 in the ML Project final evaluation.
Skills
Languages
Interests
Projects
-
kapa.ai (YC S23) Inspired Multi-Tenant RAG System
Production-grade RAG backend using FastAPI. Ran 12+ retrieval pipeline combinations before settling on Heading-Aware chunking + OpenAI embeddings + Cohere reranker + Qdrant. RAGAS scores: 0.91 Faithfulness, 0.95 Context Recall, 0.89 Context Precision.
- Implemented MCP server for seamless Claude Desktop / Cursor integration, enabling agentic tool use.
- Clean 3-layer architecture (Orchestrator → Strategy → Implementation) with dependency inversion for easy LLM/vector-store swaps.
- Enterprise features: multi-tenancy (isolated Qdrant collections), API key auth, Redis caching, PostgreSQL conversation memory, async ingestion, rate limiting, Prometheus metrics, LangSmith tracing.
- Full Docker Compose stack (PostgreSQL, Qdrant, Redis) with comprehensive unit and integration tests.
-
Credit Card Fraud Detection — Full MLOps
End-to-end fraud detection ML pipeline with class imbalance handling, SMOTE, and reproducible experiment tracking via MLflow.
- Statistical EDA to analyse class imbalance, feature distributions, and leakage risks.
- Evaluated models using ROC-AUC and precision-recall tradeoffs with MLflow experiment tracking.
- Automated ML pipeline with CI/CD and retraining workflows using DVC and Kubernetes.
-
Vehicle Insurance Purchase Prediction
ML pipeline to predict insurance purchase likelihood using tabular vehicle and customer data. Focused on probability calibration and threshold tuning to align with business risk.
- Deployed using FastAPI with CI/CD and cloud infrastructure.
-
System Threat Forecaster (Malware Prediction)
End-to-end ML pipeline to predict system infection using telemetry data. Built at IIT Madras; scored 93/100 in final evaluation.
- EDA, feature engineering, and ensemble modelling using XGBoost and LightGBM.