Charuvi
Suresh.

Problem: scanner findings arrive in different formats and need context before they can be prioritized. Approach: parse, normalize, deduplicate, triage, and report through a Python CLI.

Workflow: scan files / code → normalize → triage → report

• Parses Nmap XML, OWASP ZAP JSON, and selected source-code patterns into a shared finding model.
• Deduplicates related cross-tool findings and supports a deterministic mock analyzer by default.
• Optional Claude or local Ollama analysis; writes terminal, HTML, and JSON reports.

Problem: provide an accessible way to assess athletic movement from a recorded clip. Approach: estimate body keypoints in the browser, derive movement metrics, then apply a rule-based suitability model.

Workflow: video → pose keypoints → movement metrics → recommendations

• Processes video frames locally with MoveNet through TensorFlow.js and the pose-detection library.
• Calculates agility, balance, coordination, and reaction-time metrics and maps them to sports recommendations.
• Supports English and Hindi, local/offline data workflows, and downloadable PDF assessment reports.

Problem: bring vessel detections and local weather context into one monitoring prototype. Approach: run video frames through a configurable YOLO inference/tracking pipeline and display results in Streamlit.

Workflow: video / image → vessel inference → dashboard + weather

• Includes video, webcam, and image input paths with frame-by-frame detection and ByteTrack tracking code.
• Configures six vessel classes and includes training and inference scripts.
• Fetches current conditions and a one-day forecast from Open-Meteo. The public repository does not include trained weights or dataset images.

Problem: keep household finance tracking and personal tasks in one dashboard. Approach: a Next.js client backed by Express APIs and MongoDB.

• Tracks expenses, income, budgets, and task status through authenticated workflows.
• Provides category breakdowns, monthly trends, and income-versus-expense analytics.
• Exports financial summaries as PDFs; the repository provides source code but no live demo link.

Problem: make interview preparation repeatable and give candidates useful feedback on practice answers. Approach: a Streamlit app with OpenAI-backed question and answer workflows and local session storage.

• Generates role-specific technical and behavioral questions and supports practice and timed mock interviews.
• Returns answer feedback, strengths, improvement suggestions, and model answers.
• Persists sessions locally in SQLite rather than requiring a hosted database.

Problem: help applicants review a resume against ATS-style checks and a target role. Approach: upload a PDF through a React interface and analyze extracted text with a Flask API and OpenAI.

• Accepts PDF resume uploads and extracts resume text for analysis.
• Returns ATS-oriented feedback, skill gaps, and optional job-description matching.
• Includes targeted resume-bullet improvement suggestions and a downloadable JSON analysis.

Problem: make recipe discovery practical across search, category browsing, and saved meals. Approach: a lightweight frontend using TheMealDB API and browser storage.

• Searches and filters recipes using meal data from TheMealDB.
• Shows ingredient lists, cooking instructions, and available video links.
• Saves favorite recipes in local storage for later visits.

Problem: build a foundation for reliable event ingestion and downstream processing. Approach: a local-first API and AWS CDK infrastructure code.

• Implements an ingestion endpoint with request validation and idempotency handling.
• Includes observable pipeline metrics and infrastructure definitions for AWS services.
• Cloud deployment and later roadmap phases are not complete; the working API is local.

Problem: explore permission-aware retrieval and agent workflows for a logistics use case. Approach: prototype against synthetic company data; it is not a deployed production service.

• Uses a fictional logistics scenario and synthetic data rather than real customer information.
• Explores retrieval-augmented responses and permission-aware access patterns.
• Project remains in progress and is not deployed to Google Cloud.

Problem: help reason about Google Cloud architecture using current reference material. Approach: a multi-agent prototype backed by hybrid retrieval and cited documentation.

• Retrieves and cites Google Cloud documentation in the first project phase.
• Includes a Gemini-based agent workflow and an evaluation harness.
• Later user-facing and infrastructure features remain on the roadmap.