Machine Learning Developer | Python · Scikit-learn · LightGBM · Time Series · TensorFlow · PyTorch · OpenCV · FastAPI · Streamlit

Building forecasting models, NLP pipelines, and AI-powered developer tools.

I’m building my foundation in Machine Learning, from data preprocessing and statistical analysis to feature engineering, model training, evaluation, and deployment. I enjoy working on practical problems like time-series forecasting and NLP, experimenting with different models, and gradually turning ML experiments into usable full-stack and AI-powered tools on the side. Currently looking for an ML internship.

An intelligent agentic browser automation platform and Chrome extension that translates natural language user goals into autonomous multi-step web actions with real-time visual grounding and human-in-the-loop safety verification.

• Agentic Task Orchestration: Interprets high-level user intent using LLM reasoning, compiling
  real-time browser observations into deterministic step-by-step action sequences over low-latency
  WebSockets.

• Semantic A11y Tree Grounding: Parses live DOM structures into lightweight, noise-filtered
  Accessibility (A11y) Trees for robust cross-site element targeting without brittle CSS or XPath
  selectors.

• Human-in-the-Loop Risk Safety: Implements strict policy gating with real-time Approval Request
  prompts for sensitive transactions and seamless CAPTCHA / 2FA security challenge handover.

• Smart Bill & Document Extractor: Integrates Tesseract.js OCR and multimodal parsing to ingest
  invoices/bills and auto-fill complex forms and payment flows.

• Profile Vault & Task Scheduler: Features a contextual user identity vault with persistent background task scheduling, execution logging, and automated reminder queues.

  TypeScript Node.js Fastify WebSockets Preact TailwindCSS WXT Extension OpenAI API
Tesseract.js Zod Vitest

An autonomous, closed-loop revenue recovery pipeline and real-time dashboard for mitigating payment failures, checkout drop-offs and subscription declines with deterministic safety guardrails.

• Deterministic 5-Stage Pipeline: Orchestrates payment detection, root-cause diagnosis, policy strategy selection, channel execution, and closed-loop ledger auditing.
• Hybrid Diagnosis (Rules + LLMs): Combines instant deterministic gateway mapping with Groq & Gemini Flash fallback to classify ambiguous logs and Hinglish customer notes with zero cold start.
• Safety & Policy Guardrails: Enforces 12 registered policy rules, customer velocity limits (≤3 interventions / 24h), 48h soft-decline cooldowns, and an emergency global kill-switch.
• Multi-Channel Execution & Gateways: Dispatches idempotent Razorpay API actions and automated recovery workflows via WhatsApp, Email, SMS, and Ops escalation.
• Financial Ledger & ROI Analytics: Delivers cost-aware accounting, tracking 40% gross recovery rate, 575x ROI, 100% safety precision, and <5s average resolution latency.

Python FastAPI React TypeScript SQLAlchemy Gemini API Groq Razorpay Tailwind CSS Docker Pytest

An end-to-end Machine Learning pipeline and interactive Streamlit web dashboard for predicting retail store demand across multiple products and handling sparse time-series data.

• Automated Data Pipeline: Ingests transactional CSV logs, auto-detects encoding/schema, and standardizes daily timelines with zero-fill reindexing.
• Advanced ML Models: Uses LightGBM (Standard & Tweedie Regression) and Random Forest for regular and sparse/zero-inflated sales volumes.
• Time-Series Feature Engineering: Generates lag features (1, 7, 14, 28 days), calendar features, and 7/14/30-day rolling statistics without data leakage.
• Interactive Dashboard: Streamlit dashboard for data upload, store-item selection, forecast-horizon tuning, and prediction visualization.
• Fast Inference: Evaluates multi-step forecasts in <50ms using strict chronological time-based validation splits.

Python LightGBM Pandas Streamlit Scikit-learn NumPy Plotly Matplotlib