Megha Raj

Data Scientist with 1.3+ years of production experience building LLM-powered AI agents, multi-path RAG pipelines, and cloud-native ETL systems across a network of 6,000+ EV chargers, serving 1,700+ daily active users and processing 3.5M+ transactions. Proven track record delivering anti-hallucination safeguards, role-based access governance, real-time streaming, and ML-driven pricing in safety-critical, high-throughput environments. Skilled in Python, LangChain, LangGraph, OpenAI API, AWS, and production-grade MLOps, with a strong record of designing and shipping reliable, scalable AI systems that drive measurable business impact.

Data Scientist - ChargeMOD - Trivandrum, Kerala

(2025-05)

• Architected a modular monolith support platform with a three-path LLM reasoning pipeline (direct log-based diagnosis, hybrid BM25 + ChromaDB RAG with cross-encoder reranking, dynamic guided questioning), reducing average issue resolution time by approximately 50%.
• Built an entity-cache validation layer that verifies charger IDs and wallet balances against live API responses before LLM injection, structurally preventing hallucination in a safety-critical environment where the agent can remotely start/stop chargers.
• Delivered fuzzy charger name resolution and real-time SSE streaming for operator dashboards, enabling new support staff to handle complex charger faults independently from day one.
• Designed a conversational analytics agent for 6,000+ EV station owners using LLM tool-calling over 15-20 parameterised PostgreSQL query functions, with owner ID auto-injected from session, making cross-account data access structurally impossible.
• Implemented Redis caching, nightly batch pre-aggregation, circuit breakers, per-owner rate limiting, and prompt regression tests blocking deployments below 95% accuracy, serving 1,700+ daily active users.
• Achieved near-zero time-to-insight for energy, session, and fault analytics, eliminating manual dashboard navigation for station owners.
• Built a nightly two-level ML pricing pipeline with network-wide demand forecasting (XGBoost, Random Forest, Facebook Prophet with auto best-model selection) and per-charger pricing via an activity classifier paired with a log-space energy regressor.
• Applied KMeans behavioural clustering on 36-feature charger profiles, p95 outlier clipping, log1p transformation, and asymmetric sigmoid surge mapping within regulatory caps (+12%/-6%), delivering approximately 10% revenue uplift.
• Built a monthly churn detection system with energy-drop threshold segmentation, rolling-percentile lifecycle clustering across 5 tiers, and a proactive calling console for customer care teams.
• Delivered a Power BI management dashboard and a care-team console with call outcome logging, recovering churned revenue at a fraction of new acquisition cost.
• Designed a nightly AWS Glue + S3 Parquet + Athena serverless pipeline extracting EV transactions from MongoDB and transforming with PySpark; date partitioning reduced cloud query costs by approximately 95%.
• Built idempotent partition overwrite logic, a 48-hour lookback buffer, and schema-anomaly quarantine; Power BI auto-refreshes daily, processing 1M+ transactions nightly with 30% faster data processing.
• Built a nightly Python batch job computing gap-based uptime/downtime per charger from MongoDB logs, handling schema divergence between AC and DC chargers, contributing to approximately 20% reduction in operational downtime.

Data Science Intern - ChargeMOD - Trivandrum, Kerala

(2025-03 - 2025-05)

• Developed an EV Assistant Chatbot using OpenAI function calling, embedding-based retrieval, dynamic entity extraction, and real-time SSE streaming for route optimisation and live charger discovery; the project evolved into the production Customer Support AI Agent.

RPA Data Analyst - Software Incubator Pvt. Ltd. - Kerala

(2022-06 - 2024-05)

• Automated document extraction and data processing workflows using JavaScript, Puppeteer, and Kofax; built browser automation pipelines for structured web scraping and data indexing into MySQL databases.

M.Tech - Computer Science (Digital Image Computing) - Kerala University (2023 - 2025)

B.Tech - Computer Science and Engineering - University College of Engineering, Karyavattom (2015 - 2019)