Ashish Sharma

Krakow | ashishcred@gmail.com | +48 579 331 995

Professional Summary

AI engineering and quantitative software professional with 5+ years building Python-based risk, analytics, and production systems in financial services.
Builds applied GenAI systems end to end: document ingestion, structured LLM extraction, RAG, tool calling, vector search, access control, evaluation, and FastAPI delivery.
Currently productionizing a scenario-intelligence chatbot to accelerate economists' scenario review and sign-off workflows.
Combines model-risk domain knowledge with strong software engineering in Python, APIs, testing, and scalable deployment.

AI / LLM Engineering: RAG, tool/function calling, prompt orchestration, structured LLM outputs, embeddings, vector search, LLM evaluation, streaming chat APIs
Machine Learning: PyTorch, scikit-learn, Pandas, NumPy, time-series modelling, statistical modelling, risk modelling
Software & Data: Python, SQL, C++, R, PostgreSQL, pgvector, Pydantic, FastAPI, Flask, REST APIs, microservices
Engineering Tools: pytest, Git, GitHub Actions, Azure, Docker, JIRA, Dash

Built an evidence-first, versioned ingestion pipeline for 28 interconnected financial-model documentation PDFs, using PyMuPDF, PostgreSQL, and Pydantic-validated structured LLM outputs.
Delivered a tool-calling chatbot with a streaming CLI and FastAPI Server-Sent Events API; grounded responses in governed data and required page-, section-, and text-span evidence for every claim.
Designed a content-addressed, append-only data model across 32 tables so answers remain traceable to their source material and pipeline versions.

Building a scenario-intelligence workflow that combines a vector-database knowledge base with Python comparison scripts to generate comments on existing economic reports and accelerate economists' scenario sign-off.
Uses internally hosted Qwen 2.8 for both embeddings and chat completions; the retrieval layer ingests SharePoint deltas, chunks source material, and stores vectors in pgvector with ACL filtering enforced in SQL.
Delivered a FastAPI chat service with audit logging for tool versions, model IDs, latency, and user attribution, plus pytest coverage for tool-schema conformance and orchestration round trips.

Dec 2022 – Present | Krakow

Developed and maintained market risk models using Python, improving computational efficiency by 25%.
Migrated existing frameworks from R to Python and integrated library API calls to align with microservice architectures.
Led deployment engineering, unit testing, and scalable production implementation in partnership with modelers and IT stakeholders.

Apr 2021 – Nov 2022

Engineered and validated credit risk models (PD, LGD, EAD) in Python for a diverse portfolio.
Conducted monitoring of credit performance and key covenants, proactively mitigating risks to maintain low delinquency rates.
Recognized as "Best Credit Manager" for outstanding performance in FY22.

Aug 2016 – Mar 2021

Designed and implemented portfolio analysis models, delivering actionable insights for risk assessment.

In progress | 1-year course

Aug 2009 – Jun 2013 | GPA: 8.0/10

Major: Advanced Mathematics and Statistics

Full Stack Data Science & Machine Learning, GreyAtom (Jul 2018 – Dec 2018)
Neural Network & Deep Learning; Generative AI for Everyone; Python Programming — Coursera
Version Control with Git; Troubleshooting and Debugging Techniques — Google
JIRA Scrum Master — Coursera
Chartered Financial Analyst (CFA), Level II passed