Shivansu Soni

AI Engineer, Forward Deployment Engineer, Data Scientist, and Data Engineer with 4 years of experience building production-grade data pipelines, backend services, and ML/GenAI solutions across AWS, GCP, and Azure. Strong software engineering fundamentals and a proven track record of shipping reliable, scalable features that deliver measurable business outcomes in finance, healthcare, and retail.

Forward Deployment Engineer, Data Scientist at Fractal Analytics (2025-11 – Present)

• Re-architected a production multi-agent Cash Application and Credit Risk system from CrewAI to LangGraph, introducing deterministic state transitions, checkpointing, and human-in-the-loop gates to meet audit and explainability requirements for regulated finance workloads.
• Designed and deployed an end-to-end LLMOps framework on Langfuse — distributed tracing, versioned prompt management, A/B testing, golden evaluation datasets, and LLM-as-Judge evaluators — enabling safe prompt and model rollouts with automated regression gates.
• Built golden eval datasets covering cash application scenarios (exact / partial / one-to-many matches, unapplied cash, adversarial remittances) and credit risk decisions; implemented reference-free evaluation pipelines to grade production outputs without ground-truth labels.
• Integrated Pydantic output schemas and guardrail checks (invoice grounding, amount reconciliation, escalation rules, confidence calibration) into LangGraph agents, improving decision reliability and downstream auditability for finance compliance.
• Delivered real-time observability dashboards tracking touchless match rate, false-positive rate, confidence calibration, cost-per-decision, and p95 latency; implemented role-based user management, audit logging, and annotation queues in Langfuse for enterprise governance and continuous dataset growth from production traces.

Associate Consultant at Infosys (2024-10 – 2025-11)

• Developed and deployed an enterprise-grade GenAI-based multi-cloud automation platform for AWS, GCP, and Azure infrastructure provisioning, leveraging Claude Sonnet 3.5 with Amazon Bedrock for advanced reasoning.
• Led integration of agentic AI with Retrieval-Augmented Generation (RAG) for intelligent Terraform script generation, reducing infrastructure deployment time by 30%.
• Implemented LangGraph-based agentic architecture with autonomous agents for interpreting natural language, fetching contextual cloud data, and generating Terraform IaC files (main.tf, variables.tf); built cloud knowledge bases with Amazon Titan embeddings for semantic search.
• Automated GitOps workflows via GitHub + Jenkins pipelines for real-time deployment; created reusable Terraform modules ensuring GDPR, HIPAA, and PCI-DSS compliance. Contributed full-stack (React + Python) integration of GenAI services and prompt engineering.
• Deployed the solution on AWS Marketplace, driving a 20% increase in enterprise adoption and platform visibility.

Senior Software Engineer at Persistent Systems (2022-06 – 2024-10)

• Engineered a comprehensive multi-source data integration system unifying Confluence pages, HTML dumps, and video content through robust Python pipelines — real-time web crawling, Azure Function-based Confluence extraction, and video analysis via Azure AI Video Indexer.
• Built containerized services for data extraction, embedding generation, and storage using Azure Cognitive Services for text-to-embedding conversion and Azure Blob Storage for large-scale data management.
• Integrated Azure AI Search with an advanced prompt system and conversational memory for context-aware user interactions; delivered an end-to-end solution enabling seamless insight access across text, web, and video content through a unified intelligent interface.
• Built an advanced ETL pipeline on BigQuery within the Dremio Data Lake environment, and deployed ML models in production using the Dataiku platform, reducing deployment friction for data science teams.
• Developed a conversational AI bot using the LangChain framework with OpenAI GPT-4 for natural-language-to-SQL translation and human-readable query explanations for non-technical pharma researchers.
• Implemented the Milvus vector database for embedding storage and retrieval, and OpenAI embedding models for metadata enrichment — enabling rich semantic search across proprietary pharma data.
• Built a real-time, scalable data pipeline on GCP (BigQuery, Dataflow, Vertex AI) processing high-velocity user clickstream data for purchase intent prediction; orchestrated ingestion with Apache Airflow and used PySpark for session-level aggregation.
• Trained ML models with Scikit-learn and deployed them on Vertex AI for real-time inference, enabling dynamic marketing campaign adaptation based on user intent signals.

B.Tech in Computer Science Engineering – SRIT Jabalpur (2022)

Diploma in Computer Science Engineering – Govt. KNPC Jabalpur (2019)