12+
Years in AI and ML systems
Applied AI Leadership
Building production-ready LLM systems and agent workflows at enterprise scale.
I am Bhavani Shankar Y, an Applied AI engineering leader with 12+ years of experience delivering measurable impact across NLP, distributed systems, and cloud-native platforms.
Quick Overview
Impact
15%
ETA accuracy improvement delivered
$500K
Annual savings from cloud migration
10
Pilot customers onboarded in Doc AI
About
About
Results-focused engineering leader with deep expertise in LLM systems, planning-based agent architectures, and distributed cloud platforms. I translate research-grade concepts into reliable products that improve business performance.
Across Samsung and FourKites, I have led high-impact teams, improved product quality at scale, and delivered systems spanning conversational AI, document intelligence, and supply chain automation.
Expertise
Skills
LLM Systems
- Fine-tuning
- RAG
- Structured Output Control
- Hallucination Mitigation
- Tool-Augmented Agents
Agent Frameworks
- LangGraph
- Planner-Executor Patterns
- Workflow Synthesis
- Tool Routing
- Temporal
Platforms and Infra
- AWS (EKS)
- Azure
- Kubernetes
- Kafka
- SQS
Programming and Data
- Python
- Java
- Go
- Postgres
- Qdrant
Work
Selected Projects
Loft: Natural Language to Workflow Compiler
Designed an LLM-powered compiler that converts natural language requests into deterministic DAG execution plans using planner-executor architecture with LangGraph.
Outcome: Reduced enterprise integration time from 4 weeks to less than 0.5 days in internal evaluation.
Document Intelligence Agent
Built a document processing system for structured extraction from BOLs, invoices, and shipment updates with HITL validation and feedback loops.
Outcome: Successfully onboarded 10 pilot customers with measurable operational gains.
Conversational AI Assistant
Architected a tool-augmented assistant for enterprise structured data reasoning with model routing, observability, and low-latency response workflows.
Outcome: Improved execution accuracy through fine-tuning and prompt optimization while reducing hallucinations.
Writing
Featured Medium Articles
From Genres to Conversations: Next-Gen Movie Recommendations with LLMs
A practical look at hybrid recommendation systems combining embeddings and LLM reasoning for better personalization.
How to Make Your SLMs More Accurate
Evaluation-driven strategies to improve small language model performance through retrieval and rule-assisted controls.
Building Your GenAI Tech Stack
A blueprint for moving from quick GenAI prototypes to resilient, production-scale systems.
Read more on my Medium profile.
Resume
Resume
Download the latest resume for detailed experience, projects, publications, and education.
Open Resume (PDF)