The Role
Own end-to-end credit & fraud data science: feature engineering from raw bureau JSON ,SMS,DEVICE, scorecard / model development, Business Rule Engine (BRE) design, monitoring, and partnering with product/engineering to put rules live. You will work directly with the existing DS team,Tech,product and founders — decisions are data-backed and debated.
What You Will Own
- Build and maintain credit scorecards and models for FTB and Repeat Borrowers (Xgboost, Random forest, Support Vector Machine Models, ensemble models, challenger models).
- Engineer features from raw CRIF (or equivalent) bureau JSON — tradelines, enquiries, DPD histories, identity matches — and from raw SMS / FinBox alt-data (collections, rejections, salary, app footprint).
- Design, validate, and ship Models: hard rejects, soft flags, amount caps — with clear lift/capture
/ approval trade-offs.
- Own portfolio risk analytics: vintage / DPD / non-starter / POS bad-rate monitoring; propose tier pauses, cool-offs, and ladder-up changes.
- Build fraud signals (device, SIM/OTP, mule, ring, post-disbursal disappearance) and help prioritise the fraud PRD backlog into production.
- Partner with engineering to productionise features, rules, and models (Watchtower-style shadow underwriting, policy index, monitoring dashboards).
- Challenge and refine existing tier/ladder policy with evidence; communicate clearly to founders and business.
Required Experience
- Tenure: 5+ years overall experience in data science/analytics.
- Digital lending: Minimum 3 years hands-on in digital lending/consumer credit (NBFC, fintech lender, digital/STPL/) who has built models themselves.
- Scorecards/models: Built and deployed at least one credit scorecard (first-time borrower or repeat borrower, or combined model) into a live BRE / LOS. Should improve approval–bad-rate trade-offs from production experience.
- Bureau: Parsed and engineered features from raw bureau files (CRIF / CIBIL / Experian JSON or XML) — not only vendor-precomputed attributes.
- Non-starter models: Fraud/non-starter / First Payment default modelling experience in short-tenure lending.
- Limit Assignment: Experience with repeat-borrower ladder / limit-management policies.
- Monitoring and QC: Shadow underwriting/champion–challenger frameworks.
- Alt-data: Worked with SMS / alt-data / device / AA signals for underwriting or fraud (FinBox, similar vendors, or in-house SMS parsing).
- Stack: Strong SQL + Python (pandas, sklearn/Logistic / lightgbm/Xgboost/randomforest, statsmodels). Able to write production-quality notebooks and scripts, not just slide decks.
- Communication: Comfortable debating policy with founders/credit heads using data; owns the "show me the evidence" conversation.
Nice To Have
- Feature stores, Airflow/cron pipelines, S3 + Postgres + DynamoDB.
- Prior Experience: Prior work at a zero-to-one digital lender or STPL product.
What success looks like in 6 months
- A documented feature dictionary from raw bureau + SMS with IV/KS ranking.
- At least one new scorecard/model live with clear expected vs observed bad-rate impact.
- Non-starter / First Payment Defaults monitoring with actionable rule recommendations and clear demonstrated improvements in defaults
- Credible pushback on weak policy ideas — backed by analysis, not opinion.
Skills:- Data Science, pandas, Scikit-Learn, XGBoost and SQL