Building Production AI
& ML Systems
AI Engineer focused on production ML systems, LLM-powered applications, forecasting pipelines, and real-time analytics infrastructure. IIT Madras graduate experienced in FastAPI services, retrieval-augmented generation (RAG), time-series modeling, and scalable API-driven workflows.
Production AI Engineering
Featured Projects
End-to-end Decision Intelligence platform combining Logistic Regression with Groq-powered Llama 3.1 LLM for customer churn prediction (92% accuracy), driver explanation, and AI-generated retention strategies. Modular FastAPI backend with provider-agnostic LLM abstraction layer and scenario simulation.
📐 Architecture & Technical Highlights
- Modular FastAPI with 4 core services: DS Service, GenAI Service, Orchestrator, History Logger — clear separation of concerns
- Logistic Regression predicts churn probability from usage hours, support tickets, tenure; outputs risk tiers (High >60%, Medium 30-60%, Low <30%)
- Driver detection identifies interpretable signals: low product usage, high support volume, short tenure
- Provider-agnostic LLM factory supports Groq, OpenAI, Gemini — swappable without touching app logic
- /simulate endpoint models intervention impact (e.g., 78% → 51% risk by simulating actions)
- Batch CSV processing with per-customer explainability scores
- Interactive Streamlit dashboard for analyst consumption
Complete end-to-end pipeline for day-ahead electricity price forecasting using XGBoost. Ingests ENTSO-E API data, performs rigorous QA checks, engineers temporal features, generates LONG/SHORT/HOLD signals with confidence intervals, simulates trading PnL, and produces AI-assisted trader reports.
📐 Architecture & Technical Highlights
- 5-module pipeline: data_ingestion.py → qa_pipeline.py → forecasting.py → trading_view.py → llm_analyst.py
- QA framework: time-integrity checks, statistical outlier detection, economic sanity (demand-price relationship)
- Feature engineering: hour-of-day, weekday, lag-1h/24h/168h, rolling mean/std, momentum vs lag-24 baseline
- Trading logic: LONG if forecast > reference + band, SHORT if < reference - band, HOLD in band
- Interactive Plotly HTML output (outputs/charts/) with cumulative PnL visualization
- CSV submission with predictions + confidence intervals
- Optional Groq analyst (auto-skipped if no API key) for reproducibility with synthetic data
Real-time urban intelligence platform fusing live traffic, weather, and environmental data from multiple public APIs. Features automated schema validation, anomaly detection, statistical forecasting, and interactive multi-city geospatial visualizations refreshing on 3-minute pipeline cycles.
📐 Architecture & Technical Highlights
- Multi-source API ingestion with automated schema validation on refresh cycles
- Caching layer (3-min TTL) prevents redundant API calls and rate-limit exposure
- Anomaly detection flags statistically unusual readings; surfaces in live monitoring
- Statistical forecasting projects near-term trends for predictive alerts
- Interactive geospatial visualizations (multi-city, district-level drill-through)
- Streamlit-based workflow with auto-refresh pipeline orchestration
Other Projects
ML system for predicting cybersecurity threats using anomaly detection and classification. Feature engineering on system telemetry data to identify malicious patterns before escalation.
Full-stack influencer-brand matching platform with Vue 3 + Vite frontend and Flask backend. Async job processing via Celery, similarity-based matching algorithms, sponsor coordination workflow.
AI-powered digital health companion for seniors built with Flutter (mobile) + Flask (API). Groq-powered medical chatbot, Google Maps integration for hospital/pharmacy finder, medication reminders with voice support.
Interactive Power BI dashboard for comprehensive IPL 2024 season analysis with DAX calculations. Covers team performance, player statistics, match trends, seasonal patterns with drill-through filters.
LLM-powered desktop automation agent using Node.js and Playwright. Converts natural language commands into browser automation tasks with human approval workflow and Ollama integration.
Data-driven analysis of restaurant sales, profitability, and customer retention. Python-based analytics with exploratory data analysis, statistical modeling, and actionable business insights.