About
Lending decisions happen at machine speed now, and I work on both sides of that. I build AI systems: agentic workflows, MCP servers, Snowflake and Python pipelines. And I build the risk controls that keep them honest, from model-risk management to fair-lending and explainability testing. My field isAI / algorithmic risk in financial services (fintech lending).
Today I run risk and business intelligence atFairSquare, a multi-brand SMB lender (brands include National Funding). I'm the sole risk & BI owner across five lending brands, carrying nine-figure annual originations and the surveillance, forecasting, and executive-reporting stack behind it, built from zero.
The thread through everything I do: finding the number everyone else missed. A seven-figure-per-month write-off reconciliation gap, now fixed company-wide. A syndication expansion led under incomplete data with 3-scenario stress analysis. An eight-figure favorable swing in 2025 net charge-offs, traced to specific cohorts on the book I instrument.
My toolkit is the full credit-risk stack (PD/LGD/EAD modeling, vintage and roll-rate analysis, risk-based pricing, line management, and fair-lending compliance under ECOA/FCRA/TILA) built on Python, SQL, Snowflake, dbt, and Tableau, and increasingly automated by an AI layer I engineered: custom Claude skills, MCP integrations across Jira, dbt, Domo, and Microsoft 365, LLM-generated executive commentary, and a governed semantic layer designed for AI consumption.
Outside the day job I research how AI is rewiring financial risk. That work is published in IEEE venues with 40+ citations, featured in the Free Press Journal, and quoted in Republic World and NDTV.
Skills
AI & LLM Engineering
- Claude (custom skills, subagents, agentic workflows)
- Custom MCP server development
- Model Context Protocol (MCP) integrations
- Prompt & context engineering
- Retrieval-augmented generation (RAG)
- Structured outputs & tool use
- LLM evaluation & regression testing
- AI agent orchestration
- Semantic layers for AI consumption
- AI-assisted development (Claude Code)
AI Governance & Model Risk
- AI governance frameworks
- LLM output validation & hallucination testing
- Model-risk management (SR 11-7 style controls)
- Algorithmic bias & fairness testing
- AI explainability
- Human-in-the-loop review design
- AI use-policy & audit documentation
Risk Methods
- PD / LGD / EAD modeling
- Loss forecasting
- Vintage & roll-rate analysis
- Securitization reporting
- Risk-based pricing
- Scenario & stress analysis
- A/B testing
- Fair-lending compliance (ECOA/FCRA/TILA)
Data Engineering & BI
- SQL (Snowflake)
- Python
- R
- dbt
- ETL / ELT
- Data modeling
- Data quality testing
- Git
- Tableau
- Domo
- Power BI
- Streamlit
Certifications: CSM®, Six Sigma Green Belt, Tableau Desktop Specialist (TDS-C01), IBM Data Science.
Experience
Senior Business Intelligence & Risk Analyst · FairSquare (multi-brand SMB lender; brands include National Funding)
Jun 2024 – Present- Sole risk & BI owner across 5 lending brands. Own the weekly executive report read by 85+ leaders including the C-suite on nine-figure annual originations. Rebuilt it from scratch, cleared through one SVP and three VPs weekly, with LLM-generated macro commentary for non-analyst readers.
- Built the AI layer: custom MCP servers wiring Claude into Snowflake, dbt, Jira, Domo, and Microsoft 365; a reusable Claude skills library; and agentic workflows that turn raw stakeholder asks into scoped, executed work. Plus the governed AI-ready semantic layer with an LLM evaluation harness.
- Called the risk right under ambiguity: variance attribution traced an eight-figure favorable swing in 2025 net charge-offs to specific cohorts, and a syndication expansion built on a 3-scenario stress model with incomplete data finished well past target.
- Stopped bad numbers before the board saw them: a 2× funded-dollars overstatement, a growth misstatement, and a cross-join defect inflating a nine-figure metric by five orders of magnitude. Turned a seven-figure-per-month reporting gap into standing company protocol.
- Re-platformed the money-critical reporting: moved month-end accounting and securitization (facility tapes, SPV excess-spread) off legacy SQL Server onto Snowflake + dbt. June close reconciled loan-for-loan, and 100% of external-auditor discrepancies were root-caused and closed across two audit cycles.
- Led the BI re-platform: own Risk & Portfolio workstreams of a company-wide Domo-to-Tableau migration (400+ assets); standardized KPIs across all five brands, cutting manual reconciliation ~30%; coached 16 analysts and 3 interns in the domain.
- Shipped fast and self-directed: cut the weekly performance review from 2 days to same-day and closed 68+ tickets across six departments in H1 2026, most self-scoped straight from stakeholder conversations.
- Custom MCP servers
- Credit policy
- Securitization reporting
- Snowflake
- dbt
- Tableau
- Python
Student Analyst, Economic Research & Forecasting · Pacific Life
Jan 2024 – Jun 2024- Built 4-quarter macro forecasting models (ARIMA, LSTM, Random Forest, Prophet) on Bloomberg Terminal data; ARIMA won on out-of-sample accuracy. Delivered an interactive Streamlit forecasting dashboard and presented to senior research leadership.
- ARIMA
- LSTM
- Streamlit
- Bloomberg Terminal
Teaching Assistant, MGMT-90 · UC Irvine, Merage School of Business
Sep 2023 – Jan 2024- Supported instruction and grading for a foundational business course while completing the MS in Business Analytics.
Digital & IT Intern (Wellness Forever) · Technology Consultant (D-Sys Data Solutions) · Earlier
2021 – 2022- Analyzed 90,000+ SKUs to build and lead the impulse product category (+25% sales); cut partner-onboarding turnaround 50% through workflow digitization; built data-hygiene pipelines that reduced data errors 25%.
Risk Lab
Credit-Risk Toolkit
Three models from my day job, live in your browser. Drag the assumptions and watch the portfolio respond. It is the same math behind loss forecasting, delinquency surveillance, and pricing on a $275M+ book.
Vasicek single-factor model, the Basel II framework behind bank capital rules. Expected loss, 99.9% VaR, and the capital buffer between them.
Delinquency bucket flows: how monthly roll rates compound into 12-month charge-offs. The mechanics behind vintage & roll-rate surveillance.
What should this loan cost? Stack funding, operations, expected loss, and target margin into a break-even and target APR.
Built from scratch with no charting library: Vasicek/Basel II capital math, delinquency roll-rate mechanics, and APR construction, straight from the day job.
Projects
Property Analyzer
A real-estate underwriting engine that runs entirely in the browser: short-term + long-term rental analysis, cost-segregation tax depreciation, 3/5/10-year profitability, a recommended offer price, and a weighted 0–100 rating, all from nothing but an address (RentCast autofill, every assumption editable, live recompute). Full analyses encode into shareable URLs. Built with Next.js 15 + React 19 through AI-assisted development, with the finance engine regression-tested against the original Excel model.
Kite Algo Bot: Automated Trading System
GitHub ↗A rule-based algorithmic trading system for Indian equities (Zerodha Kite Connect) with one shared decision pipeline across backtest, paper, and live modes. An order only fires when every gate passes: signal-quality score, risk:reward, position sizing, margin, session timing, and kill switches (the same risk-controls discipline I apply professionally). Below: a real backtest replay from that pipeline.
Backtest Replay: Swing Style, 365 Days, US Large-Caps
Generated by the bot's real decision pipeline on synthetic market data over a config-driven US large-cap universe, the same code that runs paper and live modes. A demo of the system's mechanics (gates, sizing, kill switches), not a performance claim or investment advice.
Research & Writing
40+ citations
Peer-reviewed research: IEEE
Publications on predictive analytics for financial risk, AI-driven credit decisioning, comparative BI & data-analytics methods, and online transaction risk factors.
Google Scholar profile ↗Industry white paper · published by Datalogz
Migrate the Mess, or Mess Up the Migration
What the AI era demands from your semantic layer. AI agents now consume BI metrics built for human eyes, and a migration is the once-a-decade window to fix that. Centerpiece: a lending case study where three versions of “Delinquency Rate %” silently disagreed in executive reporting for over a year, and the five properties (canonical definitions, explicit grain, logic separation, rich metadata, governed ownership) that make a metric layer AI-ready. Features practitioner interviews across Domo, Tableau, Looker, Power BI, and Qlik.
Verified research identity
ORCID record
Publication and peer-review record under a verified ORCID iD.
orcid.org/0009-0001-7364-1656 ↗Free Press Journal · Apr 2026
The Analyst Building Credit Risk Infrastructure for America's Small Business Lenders
Profile on my work building the credit-risk systems behind $275M+ in SMB lending across five brands.
Read the profile ↗Republic World · Mar 2026
War-risk premiums and global oil shipping costs
Quoted as a risk expert on war-risk insurance mechanics during the West Asia conflict: premium escalation from 0.25% to 1% of hull value and its knock-on effect on energy prices.
Read the article ↗NDTV · Expert commentary
GST collections, MSME challenges & India's digital tax infrastructure
Quoted on tax-reform challenges facing MSMEs: GSTN digital infrastructure, adoption hurdles, and input-tax-credit mechanics.
Read the article ↗International financial services
Invited industry training: Ferrum Capital
Invited to deliver training on credit-risk analytics based on published research.
Recognition
Industry Advisory Board, Southeastern Louisiana University
Incoming board member, advising on analytics and industry alignment; first board term begins December 2026.
Journal peer reviewer, Library Hi Tech (Emerald Publishing)
Invited reviewer for an established, indexed, peer-reviewed journal, evaluating other researchers' work in information systems and analytics.
Conference peer review & judging
Peer reviewer, IEEE INDISCON 2026, Track 6: Signal Processing, Computing & Data Science (3 papers); reviewer, ICAIS 2026 (International Conference on Artificial Intelligence Systems); invited judge, TECHNEX 2026; judge, NMIMS University national hackathon.
Education
MS Business Analytics, UC Irvine, Merage School of Business. MBA Technology Management & BS Information Technology (Hons.), NMIMS University, Mumbai.
Certifications & compliance
CSM®, Six Sigma Green Belt, Tableau Desktop Specialist (TDS-C01), IBM Data Science. Fair-lending compliance: ECOA, FCRA, TILA.
Let's talk about risk, data, or a role.
I'm always open to conversations about risk analytics, AI governance, and BI leadership roles in financial services.
hello@koustubhsharma.com