Selected Work
Projects
Each project here was conceived from within the domain first. The architecture, data models and system logic were all defined before any code was written, drawing directly from automotive engineering and product strategy experience.
AutoPredict is a full stack predictive maintenance platform built for connected vehicle fleets. It ingests real time telemetry from vehicle TBox units across 23 data channels, processes it through six specialised machine learning models and surfaces actionable service alerts through a dealer facing React portal. The platform covers the complete loop from raw sensor data all the way through to service appointment booking.
This was not a wireframe or a concept document. The entire platform was built to production grade specifications using Claude Code and VS Code, with every component defined through detailed domain specific prompts authored from deep automotive engineering knowledge. The prompt documents covering both V1 and V2 of the platform run to over 200 pages of precise technical specification.
Domain architect and prompt engineer. Designed the entire system architecture, defined all physical signal encoding rules for 23 telemetry channels, wrote the feature engineering specifications for all six machine learning models, authored the synthetic data generation rules with physically consistent constraints, and built the AI service agent workflow from scratch using LangChain.
- Three data ingestion modes running simultaneously: real time TBox stream over MQTT and HTTP, batch CSV upload for historical data, and a synthetic data generator for model training and demos
- Signal registry as the single source of truth for all 23 channel encodings, covering scale factors, offsets, validity flags and physical range validation for every signal
- InfluxDB for high frequency time series telemetry, PostgreSQL for relational state and service records, Redis for an online feature store with 25 hour TTL per vehicle
- Kafka topic architecture with four tiers: high frequency at 10Hz for drive style signals, standard for all other telemetry, low frequency for GPS and network data, and a dedicated DTC topic with 2 year retention
- Vehicle digital twin per VIN updated in real time from five Kafka topics, stored as a single Redis document containing live state, ML predictions, active alerts and driver profile
- FastAPI REST and WebSocket API with JWT authentication, two role levels and auto generated OpenAPI documentation
- React TypeScript dealer portal covering live telemetry charts, service bay management, parts inventory tracking, driver score leaderboard and a workflow tracker for the AI service agent
- Brake wear prediction using XGBoost regression for days to replacement and a CoxPH survival model for hazard function, trained on brake stress cumulative index, harsh braking rate and high speed stop events
- Engine oil degradation using a physics based Oil Degradation Index formula covering kilometre fraction, cold start count, thermal stress, high RPM duration and fuel enrichment, with an XGBoost correction layer on top
- HV battery state of health estimation using coulomb counting per charge cycle, ARIMA trend forecasting per vehicle over 90 days, and Isolation Forest for cell voltage anomaly detection
- 12V battery failure prediction using a logistic regression and XGBoost ensemble on resting voltage trend, cranking voltage dip, parasitic drain rate and cold weather multiplier
- Tyre wear prediction using LightGBM regression and a rule based puncture detector on pressure drop rate and temperature corrected pressure imbalance across all four axles
- Fuel anomaly detection using unsupervised Isolation Forest on consumption deviation from the vehicle baseline, RPM variance and coolant rise rate
- V2 adds an LSTM with attention for 30 day failure trajectory prediction, a TCN autoencoder for 60 second window anomaly detection, and a Transformer for battery remaining useful life prediction
- Final ensemble combines tabular XGBoost predictions at 40 percent weight with LSTM sequence model predictions at 60 percent weight, classified into six failure stages from Healthy through Critical
A 10 stage LangChain powered workflow agent that moves an alert from detection through customer notification, appointment booking, parts pre ordering, workshop receipt, live progress updates, cost approval, delivery notification and post service follow up. Each stage has defined entry actions, SLA based timeout escalation and conditional transitions. Communication is handled through push notifications via FCM, SMS via Twilio and email via SendGrid.
- DTC processor decoding Base64 encoded diagnostic trouble codes from five ECU systems with severity classification and a safety critical DTC lookup table
- Feature store with an online Redis layer and an offline Parquet archive, with point in time safety enforced by a leakage checker that runs before every training job
- Eight driver archetypes with precise behavioural parameters covering urban commuter, highway cruiser, aggressive driver, eco driver, taxi fleet, delivery driver, hill region driver and elderly cautious driver
- Contextual features adjusting predictions based on road type, elevation stress, rain intensity, thermal zone and estimated vehicle load condition
- OTA event tracking with post firmware update efficiency delta measurement for BMS software updates
- Parts inventory demand model using LightGBM with safety stock calculation per part per dealer, based on historical consumption, seasonal index and supplier lead time
- Model drift monitoring using Population Stability Index with automatic retraining via Celery and a champion challenger A/B testing framework
- SHAP based explainability for every ML prediction, converting feature importance values into plain language explanations attached to each dealer alert
CCPA is a reusable implementation framework for Digital Product Passport compliance. It is not a SaaS subscription or a licensed product. It is a pre-built accelerator covering data models, compliance engines, QR infrastructure and multi-tenant architecture, deployed and configured per client engagement. Every OEM that exports batteries or electric vehicles into the EU must comply with the Battery Passport regulation by February 2027. CCPA gets them there in 6 to 8 weeks instead of 18 months.
The framework covers three regulatory regimes simultaneously: EU Annex XIII and Annex X for the Battery Regulation, China MIIT GB/T 34014 for battery traceability upload, and India CPCB Extended Producer Responsibility. No competing platform handles all three in a single deployment.
Domain SME and solution architect. Designed the full framework architecture, defined the master data hierarchy from organisation level down to individual component, authored the tri-regional compliance logic, built the go to market strategy across six geographies and three customer tiers, and defined all service lines, engagement models and revenue structure for the implementation business.
- Master data hierarchy spanning organisation, facility, plant, product and component levels with full tenant isolation per client deployment
- Compliance engine calibrated per region: EU Battery Regulation Annex XIII for material declaration, Annex X for due diligence on cobalt, lithium and nickel, MIIT for China upload automation, and CPCB for India EPR portal integration
- QR code infrastructure generating scannable passport endpoints per product unit, accessible by regulators, market surveillance authorities and end consumers
- System integrations with ERP platforms including SAP S/4HANA, Oracle and Microsoft Dynamics; PLM systems including Siemens Teamcenter and PTC Windchill; BMS systems for real time battery state data; MES for lifecycle event auto-capture; MIIT national platform API; and the CPCB EPR portal
- Carbon footprint declaration per lifecycle stage using JRC PEF-PEFCR methodology built into the compliance engine
- Supplier onboarding module with bulk data import, quality audit and cohort based rollout for Tier 1 and Tier 2 supply chains
- Extensible architecture supporting new ESPR product category modules as delegated acts land, covering textiles in 2027, electronics in 2028 and furniture in 2028
The GTM is sequenced across six markets by compliance urgency. India is the beachhead with active pipeline accounts. The EU is entered through Indian system integrator partnerships including Infosys, Wipro and TCS who win OEM contracts but need a pre-built tri-regional framework to deliver. China enters in parallel because 35 percent of Chinese battery exports go to the EU and no domestic Chinese tool generates EU Annex XIII passports. Southeast Asia, the United States and the Middle East follow from 2027 onward as the regulatory wave extends to more product categories.
- Tier 1 strategic anchor accounts at OEM level with full framework implementation, all ERP and PLM integrations, managed compliance services and a supplier rollout programme. Typical Year 1 engagement value between 1.85 and 4.45 crore rupees
- Tier 2 supply chain accounts pulled in by OEM mandate, configured in 3 weeks with lightweight data onboarding and upstream passport data export. Typical engagement between 20 and 80 lakhs
- Tier 3 institutional engagements with regulatory bodies, government frameworks and global SI partnerships for framework licensing, white label delivery and standards alignment
FinOS is a personal and household financial operating system built to handle the full complexity of Indian personal finance. Most financial apps address one slice: budgeting, or investments, or tax. FinOS covers all of it in a single platform where every module connects to every other. A salary entry flows into budget calculations, updates investment capacity, adjusts EMI headroom and feeds the financial health score automatically.
The platform was built using VS Code and Claude Code, with the same domain first approach applied here to fintech as across the other projects on this page.
- User Management covering login, profile setup and role based authentication
- Income Management for salary tracking, variable pay, freelance income and passive income streams
- Expense Management with category level logging, recurring expense detection and merchant level analysis
- Budget Planning with allocation by category against income, variance tracking and reallocation suggestions
- Financial Goals module for savings targets with projected completion dates and contribution tracking
- Investment Tracker covering mutual funds, direct equities, fixed deposits, EPF, PPF and sovereign gold bonds
- Loan Management for home loan, car loan, personal loan and credit card EMI tracking with prepayment analysis
- Subscription Tracker for OTT services, software, gym memberships and recurring services with annual cost visibility
- Notifications Engine for budget threshold alerts, bill due dates, investment maturity and goal milestones
- AI Recommendation Engine suggesting reallocation, prepayment priority and investment opportunities based on current financial position
- Forecast Engine predicting monthly cashflow for the next 12 months based on income patterns and expense history
- Couple Finance Module managing two income streams as a shared household with contribution splits and joint goal tracking
- Dashboard Analytics with charts and KPIs covering net worth, savings rate, expense breakdown and investment returns
- Automation Engine with rules and workflows for recurring transactions, auto-categorisation and scheduled transfers
- Financial Health Engine producing a composite score based on savings rate, debt to income ratio, emergency fund coverage and investment diversification
- Tax Planning module covering old and new regime comparison, Section 80C optimisation, HRA, home loan deductions and advance tax estimation
- AI Chat Assistant for conversational financial queries answered in the context of the user's own financial data
- Scenario Simulator for what if planning, showing how cashflow changes if a home loan rate rises or a large expense is brought forward
- Audit and Security module with access logs, session history and data export controls
- Reports Module generating PDF and Excel exports for monthly statements, tax summaries and investment reports