Sudhanshu Patil

AI Engineer with ~1 year owning production AI systems design → deployment as sole AI engineer for banking and enterprise clients: agentic RAG with HITL, an MCP tool server across 9 CRM modules, a 94% PII-redaction latency cut, and solo platform migrations for air-gapped systems. Backed by full-stack delivery (FastAPI, React, PostgreSQL, Redis), 8,000+ lines of production code, and shipped products with real users.

AI Solutions Developer at SimpleWorks (2025-08 – Present)

Sole AI engineer owning architecture, development, and production deployment for banking & enterprise clients

• Delivered QA-passed agentic reasoning for BFSI clients (incl. Bank of India), as measured by multi-traversal responses with up to 2 HITL levels, by building an Agentic RAG framework (Orchestrator, Intent Recogniser, Dynamic Query Splitting, Math Sandbox) on a Hub-and-Spoke architecture.
• Enabled agentic tool-use across Claude, Gemini CLI, and internal apps, as measured by 10+ production MCP tools across 9 CRM modules with configurable 40k-token memory, by architecting a cloud-hosted FastMCP server as a native query mode.
• Cut PII redaction latency by 94% (8s → 0.5s) by replacing a spaCy transformer with fine-tuned NER DistilBERT and adding triple-verification guardrails that block excluded fields from every LLM payload.
• Unblocked GPT-5-series adoption, as measured by zero-rollback upgrades on QA and PSB Production, by solo-leading Python 3.11→3.13 + LlamaIndex migrations (Azure AI Foundry), incl. offline wheel packaging for air-gapped banking systems.
• Sustained production reliability as a team of one, as measured by 8,000+ lines of production code, 5+ major SQL fixes, and 20+ QA defects resolved, by owning architecture → development → deployment.

Associate Data Scientist (Intern) at Novartis (2025-01 – 2025-07)

Awarded full-time Pre-Placement Offer (PPO) for performance — declined to join SimpleWorks

• Kept sensitive clinical data fully on-prem, as measured by R-code responses of OSS-comparable quality with zero data egress, by building an on-prem LLM chatbot trained on internal R-coding documentation.
• Eliminated manual intervention for 60+ Data Managers, by shipping VBA + Office Scripts automations (auto-filing, input guardrails) and replacing SAS reporting with an interactive RShiny dashboard.

B.E. (Hons.) in Electronics & Communication Engineering + Minor in Computing & Intelligence – BITS Pilani, Hyderabad Campus (2021-10 – 2025-07)