Available for new opportunities
Abhishek Kumar
Lead Generative AI Engineer
LLM Platforms · Multi-Agent Systems · RAG at Scale
10+ years across software, data, and ML engineering — the last 3+ leading the design and delivery of production GenAI platforms for two global pharma organizations, Eli Lilly and Bristol Myers Squibb. I own end-to-end architecture of LLM applications: agentic systems, RAG, evaluation and observability, and GPU inference on AWS and Kubernetes.
- 10+Years across software, data & ML engineering
- 3+Years leading production GenAI delivery
- 60–70%LLM inference cost reduction
- 3M→3KCompounds triaged by agent workflow
Experience
I have been working across GenAI, data systems, and cloud architecture. Here is a timeline of my journey.
Apr 2025 - Present
Lead GenAI Engineer · Syngene International | Bristol Myers Squibb
Bengaluru · Leading a team of 5 engineers building GenAI platforms for BMS drug-discovery R&D
- Cut LLM inference infrastructure costs 60–70% by migrating always-on SageMaker endpoints to autoscaled GPU serving (A100/A10) on EKS with queue-based workload management and event-driven execution, while keeping production latency predictable.
- Compound Library Design: built and shipped a multi-agent LLM platform (LangGraph) used by BMS computational chemists for iterative compound analysis — agent routing with governed tool access, short and long-term memory, and LangSmith-based evaluation for audit trails in a regulated research environment; onboarded 5 project teams.
- High-Throughput Screening: designed an LLM-assisted triage workflow that screened 3M compounds down to 3,000 high-confidence hits using a supervisor-orchestrated agent debate loop (advocate / critic / hypothesis agents) with human-in-the-loop calibration — eliminating ~5 months of manual review effort.
- ChatEDA: delivered a natural-language analytics platform over structured R&D datasets, translating user intent into validated transformations, visualizations, and RDKit cheminformatics operations with guardrailed tool invocation.
- Standardized model-container deployment patterns adopted across agent workflows, shortening iteration cycles for new model rollouts.
- LangGraph
- FastAPI
- Redis
- AWS EKS
- LangSmith
Nov 2023 - Mar 2025
LLM Engineer · Eli Lilly and Company
Bengaluru · Led a team of 4 engineers; architected the enterprise GenAI RAG platform serving multiple business units
- Architected an enterprise RAG platform (LangChain, AWS Bedrock, EKS) and defined the org's standards for embedding-model selection, retrieval strategy, and deployment patterns — adopted by 5 business units, driving org-wide GenAI adoption.
- Sharply reduced retrieval infrastructure costs while improving answer relevance by introducing hybrid semantic + keyword retrieval (Pinecone + Elasticsearch) and a reusable enterprise memory layer shared across business units.
- Implemented role-based, user-aware document access control in the retrieval layer, enabling compliant knowledge access across departments.
- Built automated evaluation and governance for RAG quality — response grading, regression testing, drift monitoring — cutting manual validation effort ~40%.
- Deployed open-source Hugging Face models as containerized microservices behind a model-abstraction layer, reducing vendor lock-in and enabling rapid onboarding of new models.
- Extended the platform into domain-specific agents (SQL, JIRA) with agent-to-agent orchestration, automating internal decision workflows.
- LangChain
- AWS Bedrock
- Pinecone
- Elasticsearch
- TruLens
Sep 2020 - Jun 2023
Data Engineer · Biofourmis
Real-time streaming platforms for medical and IoT sensor data at scale
- Led development of a real-time streaming platform for medical and IoT sensor data (PySpark, AWS Lambda, Redshift, EKS), supporting continuous patient health monitoring at scale.
- Built data-quality and ETL pipelines across relational, NoSQL, and file-based sources, cutting manual validation overhead and improving reliability of downstream analytics.
- Mentored engineers on distributed processing patterns and production best practices; resolved production issues spanning memory optimization and Airflow orchestration.
- PySpark
- AWS Redshift
- Lambda
- Airflow
- PostgreSQL
Aug 2019 - Aug 2020
Software Engineer · Maple Labs
Client: Data Capital Management, US
- Built data pipelines and backtesting workflows for stock-market analytics; refactored a monolith into Python/AWS microservices, improving scalability and deployment agility.
- Developed the Snappy Flow "Poller" application (Python/Django backend, Next.js/Redux frontend) with Jenkins CI/CD on AWS.
- Python
- Django
- React
- Next.js
- AWS
Oct 2015 - Aug 2019
Application Development Analyst · Accenture
MLOps and data governance for enterprise analytics
- Deployed ML models (Python, R) as containerized REST APIs within an MLOps framework for Data Science Lab initiatives; applied Random Forest and time-series modeling for enterprise predictive use cases.
- Engineered GDPR-compliance data-governance pipelines (IDAG platform) and data-ingestion services for NBN Australia's GDSS project (Flask, PostgreSQL).
- Python
- Django
- Docker
- PySpark
- PostgreSQL
Technical Skills
Generative AI / LLM
- LLMs
- Agentic AI
- Multi-Agent Orchestration
- LangGraph
- LangChain
- vLLM
- TruLens
- AWS Bedrock
- OpenAI GPT
- Anthropic Claude
- Hugging Face Transformers
- MCP
- Agent-to-Agent (A2A)
- RAG
- Prompt Engineering
- LangSmith
- Guardrails
Retrieval & Vector Search
- Pinecone
- Elasticsearch
- Hybrid Retrieval
- Embeddings
- Semantic Search
- Embedding Benchmarking
Languages & Frameworks
- Python
- SQL
- PySpark
- FastAPI
- FastMCP
- Pandas
- Django
- Flask
Cloud & Infrastructure
- AWS EKS
- Lambda
- S3
- SageMaker
- Redshift
- Kubernetes
- Docker
- Redis
- GPU Inference (A100/A10)
- Autoscaling
- Jenkins
- Airflow
- MLOps
Databases
- PostgreSQL
- MongoDB
- Redis
Education
June 2011 - June 2015
B.E. in Information Science & Engineering · Dayananda Sagar College of Engineering
VTU University · Bengaluru, India
Interested in architecting autonomous multi-agent systems or scaling high-throughput LLM pipelines?
Currently collaborating on cutting-edge generative discovery workflows, FastMCP interfaces, and distributed agent platforms.