Priyank Vashisht – AI Engineer | Public Profile on Dr.Job

About Me

AI Engineer with experience in developing and deploying agentic document intelligence pipelines
and Retrieval-Augmented Generation (RAG) platforms on Azure Databricks. Proven ability to fine-tune large
language models, e…

Experience

AI Engineer

The LCF Group, Noida, India · Noida , Uttar Pradesh , India

Jul 2025 - Present · 11 months

Developed an agentic document intelligence pipeline on Azure Databricks, leveraging Gemini-powered OCR and entity extraction to process over 800 bank statements daily across 6+ banking formats, achieving a manual correction rate below 5% and accelerating credit decisioning while enhancing loan approval quality. Engineered a Medallion Lakehouse architecture on Azure Databricks using Delta Live Tables and Auto Loader; implemented the BSI Metrics Engine on Gold datasets with the Mosaic AI Agent Framework, enabling real-time financial health scoring and compliance analytics. Optimized real-time event delivery through a Redis-cached serving layer and Delta Sharing APIs, successfully maintaining 100 ms latency Service Level Agreements (SLAs) across 3+ downstream services, handling over 80,000 monthly events. Constructed a multi-provider Retrieval-Augmented Generation (RAG) platform (LangChain + Mosaic AI on Azure Databricks) featuring hybrid retrieval and sub-200 ms response times; orchestrated agents via MCP-driven control with constitutional AI guardrails for comprehensive MCA tracking and compliance-aware automation.

ML Intern

Austere Systems Limited, Gurgaon, India · Gurgaon , Haryana , India

Feb 2025 - Jun 2026 · 1 year 4 months

Fine-tuned Mistral-7B-Instruct-v0.2 using QLoRA on a specialized corpus of First Information Reports (FIRs), charge sheets, and court judgments; developed the data pipeline from raw ingestion to instruction-tuned training pairs, resulting in a 40% acceleration in case processing and over 55% precision in identifying corruption patterns. Engineered a Hinglish-aware preprocessing pipeline for code-switched FIRs, incorporating custom tokenization and transliteration normalization, which improved accuracy by 18% compared to the multilingual baseline; validated performance against more than 500 annotated passages.

Skills

Python SQL JavaScript Azure Databricks Delta Lake Delta Live Tables Auto Loader Mosaic AI Agent Framework Apache Spark Structured Streaming Cosmos DB Azure Service Bus Azure Event Hubs Azure Functions Azure AI Foundry Unity Catalog Claude API Anthropic MCP Multi-Agent Orchestration Multi-Provider LLM Orchestration Retrieval-Augmented Generation (RAG) Vector Search (Qdrant) LangChain Prompt Engineering Constitutional AI Guardrails QLoRA Mistral-7B Fine-Tuning HuggingFace Transformers Evaluation Pipelines FastAPI Redis Docker Postgres Nginx Azure DevOps Fly.io WebSockets Machine Learning Fundamentals Data Analysis APIs LLMs AI Agents Data Ingestion ETL Problem Solving Attention to Detail Eagerness to Learn Leadership Skills Machine Learning Cloud Services Conversational AI