Shama Medam

Results-driven Data Scientist and Generative AI Engineer with 4+ years of experience at Accenture, delivering AI-powered automation solutions for global clients, including Google and Fidelity International (FIL). Expertise spans end-to-end ML/NLP workflows, LLM integration using OpenAI GPT-4 and AWS Bedrock, RAG and GraphRAG-based system design, and agentic AI development with LangGraph. Proficient in building and deploying scalable Generative AI solutions with both closed-source (OpenAI) and open-source large language models, leveraging vector databases and PostgreSQL for enterprise knowledge retrieval.

Proven track record of reducing manual effort by up to 97% through intelligent automation. Holds MTech in AI & ML (NLP Specialisation) from BITS Pilani (GPA: 8.22/10). Recognised with the Innovator Award and Tech Star Award.

Experienced in Agile delivery using JIRA and Confluence.

Enterprise AI-Powered Conversational Intelligence Platform (AWS Bedrock) | Apr 2026 – Present

• Built a scalable, enterprise-grade AI question-answering and retrieval system on AWS Bedrock, enabling natural-language querying across structured and unstructured data with grounded, cited responses at low latency.

• Developed LLM-driven query understanding pipelines — intent classification, entity recognition, query decomposition, and time-range extraction — using Bedrock foundation models and OpenAI-compatible prompting patterns.

• Designed a RAG architecture with Titan Embeddings and Amazon OpenSearch as the vector database, enabling scalable semantic search over enterprise knowledge stores; PostgreSQL was used for structured metadata and audit storage.

• Implemented concurrent multi-source retrieval with Reciprocal Rank Fusion (RRF) for improved recall and ranking across SharePoint, internal systems, and external data — aligned with GraphRAG-style multi-hop reasoning patterns.

• Built LLM-based orchestration and dynamic tool selection via function calling, routing queries to the appropriate retrieval mechanism across a distributed knowledge graph.

• Enabled streaming responses (SSE), inline citations, and automated follow-up question suggestions; optimised serverless performance using async processing, batching, and concurrency controls.

• Applied production best practices in chunking, deduplication, metadata filtering, and observability for a reliable, auditable system.

FIL Scaled AI — Financial Report Validator (Agentic AI | LangGraph) | Jan 2026 – Apr 2026

• Architected an end-to-end multi-agent validation system using LangGraph to automate compliance analysis of complex financial fund reports, orchestrating document ingestion, rule evaluation, and violation reporting within a stateful agentic graph.

• Designed the Document Extraction Agent to parse and structure multi-format financial reports, with an Extraction Evaluation node that automatically reroutes low-confidence documents for re-extraction before downstream processing.

• Built the Rules System Agent as the central orchestration node, dynamically dispatching section-rule pairs to specialised Utility Agents for granular compliance checks across formatting, semantic, and regulatory dimensions.

• Developed modular Utility Agents powered by LangGraph nodes with targeted prompt engineering strategies, enabling parallel, high-precision violation detection while minimising false positives.

• Implemented a Violation Aggregation layer and Annotation Engine that programmatically marks violations inline on source PDFs, delivering audit-ready, evidence-linked reports.

• Translated regulatory and business rules into machine-interpretable LangGraph tool schemas; delivered within an Agile sprint framework tracked in JIRA and documented in Confluence.

FIL ISS — GenAI Self-Learning Engine for Email Classification | Jun 2025 – Dec 2025

• Designed and implemented a governed self-learning framework combining RAG-based retrieval, embedding similarity checks, and trust-threshold validation to iteratively improve classification accuracy using user feedback.

• Built a trust-based validation engine to score incoming feedback before model learning, preventing noisy or adversarial inputs from degrading production model performance.

• Architected separation between the Feedback Event Store and Feedback Gold Store for auditability and governance, with PostgreSQL-backed metadata storage and full traceability across the feedback lifecycle.

• Ensured data protection compliance and transparency across all feedback processing stages.

MTech — Artificial Intelligence & Machine Learning (NLP Specialisation) | BITS Pilani, Rajasthan | Nov 2022 – Nov 2024 | GPA: 8.22/10

Coursework: Artificial Intelligence, Machine Learning, Deep Learning, Natural Language Processing

BE — Computer Science & Engineering | Sona College of Technology, Salem | Jun 2017 – May 2021 | GPA: 9.2/10