CAS (Chemical Abstracts Service, ACS)
AI and Software Development Intern
Enterprise AI & Cloud Applications
At Chemical Abstracts Service (CAS), a division of the American Chemical Society, engineered full-stack agentic AI and machine learning systems serving scientific product teams. Architected serverless cloud data pipelines across Amazon S3 and Amazon DynamoDB, deployed production Retrieval-Augmented Generation (RAG) graphs with specialized vector databases, and instituted C4 component modeling alongside automated GitLab trunk-based deployment pipelines.
Technical Workstreams & Architecture
3 Core AreasAgentic AI & Vectorized RAG Systems
Constructed multi-turn agentic workflows using specialized LLM agents and Claude developer tooling to synthesize unstructured chemical literature, patents, and property registries with sub-second retrieval benchmarks.
- ―Deployed hybrid dense/BM25 retrieval pipelines connected to specialized vector databases.
- ―Formulated deterministic guardrails preventing model hallucination in scientific queries.
- ―Integrated continuous fine-tuning loops to improve entity extraction recall on chemical nomenclature.
Serverless Cloud Data Tier & State Persistence
Designed high-throughput, low-latency cloud infrastructure leveraging Amazon Web Services to handle persistent agent memory, stakeholder request queues, and large-scale document transformations.
- ―Engineered partition key schemas on Amazon DynamoDB for sub-10ms session cache lookups.
- ―Provisioned encrypted Amazon S3 pipelines for high-volume scientific document ingestion and pre-processing.
- ―Implemented least-privilege IAM policies and automated resource monitoring.
C4 Architecture & Trunk-Based Deployment Safety
Authored formal C2 and C4 architectural diagrams to map subsystem boundaries across enterprise engineering teams, establishing clean API contracts and continuous deployment velocity.
- ―Structured modular service contracts to eliminate cross-team coordination bottlenecks.
- ―Instituted automated GitLab CI/CD regression gates enforcing zero typecheck regressions before merge.
- ―Adopted trunk-based development practices reducing release branch divergence.
Verified Deliverables & Outcomes
Empirical EvidenceBenchmarked across scientific paper and patent chunk lookups.
Zero regression merges through automated GitLab deployment gates.
Documented component boundaries aligned with enterprise standards.
- ―Engineered agent-based AI and ML full-stack applications for product stakeholders on AWS S3 and DynamoDB, and deployed agentic AI workflows using vector databases, RAG and model training.
- ―Streamlined deployment pipelines with trunk-based development through GitLab, using Claude tools and agents.
- ―Contributed to C2/C4 architecture designs, improving cross-functional development efficiency and quality.
Technologies & Tooling
Specific tools, libraries, and frameworks leveraged during this tenure:
Engineering Methodologies
Operational practices and systemic disciplines enforced:
“Enterprise AI is only as trustworthy as the data boundary beneath it. Designing clean DynamoDB partition keys and modeling strict C4 contracts proved far more critical to production reliability than the model weights alone.”