Ten projects
The work
Pipelines, models and LLM systems. Each one states what was measured and what was not.
Data Engineering Data Engineering Real-Time Quick-Commerce Operations Lakehouse
A dark store promising 10-minute delivery stocks out in minutes. Orders stream in every second and the ops dashboard has to show the problem now, not after the next batch window.Kafka / RedpandaSpark Structured StreamingDelta LakeDatabricks3 Kafka topics, 12 partitionsRead case study →Data Engineering UPI Payments Compliance Lakehouse
Payments at PhonePe scale: every transaction must land, get cleaned, get PII-masked and be risk-queryable in minutes — with an audit trail a regulator would accept.AWS S3LambdaRDSDynamoDB5,074,342 UPI events generatedRead case study →Data Engineering SaaS Revenue Metrics Platform
A SaaS CFO gets three different ARR numbers from three teams. This is the single pipeline everyone trusts — 511,089 invoice lines, deliberately dirty, transformed into one governed star schema.SnowflakedbtAirflowSQL511,089 invoice lines transformedRead case study →Data Science / ML QuickCommerce Demand Intelligence & Personalization Engine
A 10-minute delivery promise only works if each dark store holds the right stock — not so much that it spoils, not so little that orders are lost — and if promotions reach customers who will actually use them.PythonXGBoostscikit-learnPySpark1.3M order lines simulatedRead case study →Data Science / ML Credit Default Risk Prediction
At the moment of loan application: given everything known about a customer, how likely are they to default? The real output isn't a yes/no label but a ranking that drives approval, credit limit and risk-based pricing.PythonPyTorchpandasscikit-learnROC-AUC primary metric, chosen deliberatelyRead case study →- Linked pair — analytics written by hand, then an LLM that writes the analytics for you
Data Science / ML Customer Spending & Risk Analytics
Banks generate millions of transactions. Buried in them: who the most valuable customers are, whose spending is behaving abnormally, and which customers are genuinely risky.PostgreSQLAdvanced SQLPythonpandasSQL-first aggregation in Postgres, not pandasRead case study →AI Engineering Fine-tuning Gemma 3 4B for Text-to-SQL with QLoRA
Turn natural-language questions into SQL for a banking database, using a model small enough to fine-tune on a free Colab T4.Gemma 3PEFT / LoRAQLoRAHugging Face TransformersGemma 3 4B fine-tuned on a free T4Read case study →
AI Engineering Financial Text Summarization & Insight Generation
Analysts read hundreds of pages of annual reports and credit reviews. A full report exceeds any transformer's context window, and summarizing blindly buries the risk disclosures the reader actually opened it for.Hugging Face TransformersRAGPythonROUGEROUGE quantitative summary evaluationRead case study →AI Engineering WQ Auto Alphas — LLM-Driven Alpha Research
Alpha research is a search problem: generate candidate expressions, simulate them, keep what survives. Doing that by hand is slow and biased toward whatever you thought of last.PythonGemini APIWorldQuant BRAIN APISQLite RAGRAG SQLite knowledge bank over papers + examplesRead case study →