Pallaw Kumar

Senior AI Engineer with 11 years in software engineering, including 3.5+ years building GenAI systems, combining production RAG and agentic workflows with Python and Azure. Brings banking/financial-services integration experience and led clinical AI delivery that reduced coding case-handling time by 40–45%. Open to relocation.

Senior Data Scientist (GenAl) at FRACTAL ANALYTICS (2025-12 – Present)

• Led four engineers from requirements and technical design through deployment of a microservices-based multi-agent platform using Docker, Kubernetes autoscaling and GitHub Actions CI/CD.
• Designed Azure AI Search hybrid retrieval with BM25, dense embeddings and cross-encoder reranking; improved retrieval accuracy ~15% and kept p95 latency below one second.
• Built Databricks/Spark ingestion and incremental indexing across SAP ERP and SQL Server, processing ~170,000 clinical codes per quarter.
• Implemented LangGraph retrieval, validation and reconciliation workflows with Pydantic, human escalation, Cosmos DB persistence and Redis caching.
• Added RAGAS evaluation and regression gates; tracked faithfulness, relevancy, context precision/recall and grounding in Arize AI. Load tests exposed connection-pool limits.
• Shipped citation-backed CPT recommendations for 1,000+ medical coders, reducing case-handling time 40–45% vs manual search and coding; added prompt-injection/data-leakage controls.

Senior Data Scientist (GenAl) at FRACTAL ANALYTICS (2025-02 – 2025-12)

• Delivered an Azure OpenAI/LangChain assistant for the Chief Data Office, surfacing grounded KPI insights across 250+ digital products.
• Modelled 3,500+ Neo4j KPI relationships for multi-hop dependency and downstream-impact analysis, reducing manual correlation investigation from hours to seconds.
• Built validated Text-to-Cypher with 1.2–2.5-second latency; exposed 15+ FastAPI MCP tool servers adopted by 5+ internal teams.
• Built KPI forecasting workflows using five years of historical data, owning feature engineering and model retraining; fine-tuned an LLM to summarize KPI metrics and support analysis.
• Used RAGAS and LangSmith/LangFuse traces, token budgets, dynamic routing, A/B and canary testing to reduce LLM inference cost ~35%.

Technical Project Lead (GenAl) at HCL TECHNOLOGIES (2023-05 – 2025-02)

• Led LangChain/Azure OpenAI Text-to-SQL design across 20–30 ERP/supply-chain tables; evaluated 100+ benchmark queries.
• Reached 90%+ schema/table selection accuracy and 85%+ SQL correctness using schema-aware retrieval, Pydantic outputs and deterministic validation.
• Applied SQLGlot AST policies, read-only roles, table allowlists, parameterization, timeouts and result limits; blocked 100% of unsafe operations tested.
• Owned design docs and Pytest regression checks; coordinated stakeholders and reduced invalid completions ~25% over six months.

Technical Lead at HCL TECHNOLOGIES (2021-08 – 2023-04)

• Architected a Python deployment assistant for 45+ teams and ~800 developers, supporting 1,200+ monthly self-service deployments and automating ~50% of manual release work.
• Built Rasa/AllenNLP multi-turn intent flows, reaching ~93% recognition accuracy and reducing configuration-related release failures ~35%.
• Integrated Docker, Kubernetes, Terraform, Key Vault and vulnerability checks, cutting release turnaround from 45 minutes to under 10.

Senior Python Developer at WIPRO TECHNOLOGIES (2019-11 – 2021-07)

• Developed Python NLP entity and synonym pipelines for chemical research retrieval using spaCy, NLTK and AllenNLP, reaching approximately 89% NER F1 and improving top-five recall by about 28% in the referenced evaluation.
• Scaled Apache Solr semantic search across a 1M+ document index and used Redis caching to achieve approximately 15 ms median query latency.
• Designed a JanusGraph chemical ontology spanning 10M+ nodes and 40M+ edges, with parallel bulk ingestion processing 500K+ records per hour.

Senior Integration Engineer at WIPRO TECHNOLOGIES (2015-10 – 2019-12)

• Integrated 10+ banking systems using TIBCO BusinessWorks, IBM MQ and JMS, processing 5M+ daily transactions with approximately 99.9% message-delivery reliability.
• Built SOA/REST/SOAP integrations, XML/JSON transformations, retries and dead-letter queues, reducing production incident-resolution time by approximately 30%.
• Standardized IBM UrbanCode Deploy pipelines for TIBCO BW and IBM MQ across DEV/SIT/UAT/PROD, reducing manual deployment effort by about 60%.

Masters in Software Engineering – BITS Pilani (2019)