FIELD NOTES / PRODUCTION SYSTEMS

Writing

Technical case studies from systems that had to survive outside a notebook.
015 / PORTFOLIO

Making this website: a résumé, a retro operating system and the space between

How I designed and built this portfolio as three views of the same person, and what broke while turning a visual idea into a usable, maintainable website.

INFRA
READ ARTICLE · 11 MIN
014 / PRICE-GOA

Goa pricing: replacing a weekly file ritual with a 30-minute system

How I automated Goa base fare with multi-source inputs, hierarchical fallbacks, common publication and a narrow emergency path.

MLPRICINGSTATISTICAL
READ ARTICLE · 12 MIN
013 / DS-NEMO-API

Nemo API: taking uncached p99 latency from 150 ms to 17 ms

How connection ownership, a read-oriented schema and a query shaped around the endpoint turned Nemo into a much faster service without adding a cache.

GEN AIML
READ ARTICLE · 11 MIN
012 / AIB-TANISHA

AIB-Tanisha: asking for app reviews without manipulating trust

A policy-aware voice system designed to correct review-selection bias while respecting ambiguity, complaints and the customer's right to say no.

GEN AI
READ ARTICLE · 13 MIN
011 / PRICE-VARIANT

Pricing release: moving base fare from model group to car variant

How I moved to variant-level acquisition cost without allowing the real Zoomcar fleet mix to create an accidental platform-wide price drop.

MLPRICINGSTATISTICAL
READ ARTICLE · 10 MIN
010 / PRICE-FLEX

Pricing release: designing limited and unlimited kilometre packages

How booking-distance distributions and elasticity scenarios became packages that balanced conversion, cancellations, GBV and host wear.

MLPRICINGSTATISTICAL
READ ARTICLE · 12 MIN
009 / PRICE-ASSET

Pricing release: adding an auditable asset-level multiplier

How I added granular per-car control while keeping business decision logic separate from a safe, auditable execution layer.

MLPRICINGSTATISTICAL
READ ARTICLE · 8 MIN
008 / DS-NEMO

DS-Nemo: a review summarisation system

A log of the review-summarisation problem and how I tackled it.

GEN AIML
READ ARTICLE · 10 MIN
007 / DS-OSIRIS

Osiris: my best failure in classical machine learning

More than 100 modelling runs produced a strong booking-intent ranker—and experiments revealed that we had optimised the question the requester asked, not the decision the business actually needed.

ML
READ ARTICLE · 18 MIN
006 / DS-PIXEL

Pixel: evaluating generative image enhancement for a marketplace

From exploratory image edits to an end-to-end generation, quality-gating and delivery pipeline designed to improve presentation without changing the car.

GEN AIML
READ ARTICLE · 14 MIN
005 / DS-CIPHER

Cipher: evidence-grounded automation for support decisions

A clear account of the decision problem, evidence pipeline, human adjudication, selective automation evaluation and duplicate-safe execution behind an LLM-assisted support system.

GEN AI
READ ARTICLE · 13 MIN
004 / DS-WATCHDOG

Watchdog: building a stateful pricing anomaly evaluator

A plain-language and technical guide to detecting unusual daily movement in conversion, GBV and pricing without confusing seasonality or broken data for a business problem.

MLSTATISTICAL
READ ARTICLE · 12 MIN
003 / DS-PIGEON

Pigeon: standardising analytical email as an operational system

How I replaced scattered one-off mailers with a gated, previewable reporting layer that makes analytical state understandable.

INFRASTATISTICAL
READ ARTICLE · 10 MIN
002 / PRICING

Pricing: operating a multi-layer marketplace pricing system

How base fares, duration products, dynamic multipliers, host recommendations, asset controls and change management fit into one production ecosystem.

MLPRICINGSTATISTICAL
READ ARTICLE · 16 MIN
001 / DS-ALERTS

DS-Alerts: one reliability language for data science and data engineering

How execution health, infrastructure failures, dependency access and database integrity became one central, routed alerting service across DS–DE.

INFRASTATISTICAL
READ ARTICLE · 12 MIN