I built Cyber-Sphere as an AI cybersecurity defense platform with Next.js, Genkit, and Firebase. Its README frames the project as supporting SDG-16 through accessible security tools, and describes a set of AI security features alongside a panic response protocol.
My Cyber-Sphere platform supports SDG-16 (Peace, Justice, and Strong Institutions) by providing accessible security tools.
- My Cyber-Sphere AI security tools are Link Scanning, Message Detection, QR Analysis, Deepfake Verification, and Malware Detection.
My Cyber-Sphere Panic Protocol provides real-time geolocation dispatch to emergency contacts.
My Cyber-Sphere UI supports English and Hindi; high-contrast, enterprise-grade light and dark themes.
My Cyber-Sphere Unique Neural Logo is an animated identity for a professional feel.
My Cyber-Sphere Firebase configuration is already included in src/firebase/config.ts.
I need to ensure my Firebase project's Authentication and Firestore are enabled in the Firebase Console for Cyber-Sphere.
My Cyber-Sphere Vercel deployment requires the environment variable GOOGLE_GENAI_API_KEY: Your Google AI / Gemini API Key.
git init
git add .
git commit -m "Initial deploy: Cyber-Sphere Security Mesh"
git branch -M main
git remote add origin https://github.com/YOUR_USERNAME/YOUR_REPO_NAME.git
git push -u origin main
My Cyber-Sphere deployment instructions say: Click Import next to your new GitHub repository.
My Cyber-Sphere deployment instructions mark the Vercel environment-variable setup as critical.
My README's deployment flow is to create a GitHub repository, push the project from a terminal, import that repository into Vercel, configure the required environment variable, and deploy. It also says Firebase Authentication and Firestore should be enabled. I do not have a repository URL, deployment URL, Firebase project identifier, or API key value recorded here, and I have not recorded whether deployment or Firebase setup is complete.
I built mini-content-engine as a JavaScript project: an Express job API backed by PostgreSQL, with a no-framework frontend. The API accepts product briefs and optional reference images, tracks queued work, and returns a deliberate placeholder image result; Gemini can refine the image prompt when configured.
The project description says: “An Express + PostgreSQL job API with a no-framework frontend.”
The POST /generate endpoint accepts multipart form with productName, description, optional image``.
The POST /generate endpoint returns 202 { id, status }.
The GET /jobs/:id endpoint returns the job, including its status, prompt, and resultUrl when complete.
The GET /health endpoint reports application and database mode.
The README says: “Leaving it unset starts a clearly labeled in-memory demo mode.”
The README says: “Without it, the local prompt builder remains fully functional.” This refers to the optional GEMINI_API_KEY setting.
The README says: “The image-generation function is intentionally mocked as required by the brief.”
To move beyond the submission version, I can follow the README instruction: “Replace placeholderImage() in server.js with a provider call when moving beyond the submission version.” No particular image provider is named.
For local setup, the README says: “Run npm start and open http://localhost:3000.” It also instructs me to copy .env.example to .env, configure DATABASE_URL for PostgreSQL or leave it unset for in-memory demo mode, and run npm install.
For Render or Railway, the README instructs me to push the directory as a Git repository or configure it as the service root, create a PostgreSQL database and set its connection string as DATABASE_URL, and create a Node web service with build command npm install and start command npm start.
The README dates the repository start and last update to 2026-07.
A repo on @Gradientblack9's GitHub.
Hacker House Goa 2026 is a personalized ID card generator where users upload a photo, add details, customize their card, preview it, download it, and share it on X.
I started the project in 2026-08 and it was last updated in 2026-08.
The project uses HTML.
The project description uses the emojis 🌴💻🪪.
My Pinku_Sharma_Portfolio repository is a student web development project. Its project summary describes modern, brutalist-style websites, while the README uses a different description; I keep the distinction explicit rather than treating the README as confirmation of the brutalist style.
I started my Pinku_Sharma_Portfolio project in 2026-01.
My README describes the portfolio as: Student web developer creating modern, websites with HTML and CSS.
The project summary says “brutalist-style,” but that detail is absent from the README description. Both descriptions identify the project with modern websites and HTML and CSS.
I built SENTINEL as an interactive prototype based on the supplied SIH26165 PRD.
I can run the prototype with these requirements: Node.js 18+; run node server.mjs and open http://<redacted>:4173; no package installation or API key is required.
The dashboard features are: Command center, date and site filters, priority queue, activity chart, rule heatmap.
I can manually analyze reports and import CSV files with validation and duplicate detection.
The analysis and review features are: Exact source highlights, advisory consequences and controls, five-factor risk score.
The retrieval and recurrence features are: Lexical related-report retrieval, activity-based recurrence radar and interactive graph.
The HSE review features are: HSE confirmation, correction and escalation queue, with local audit history.
The dataset and export features are: CSV export and resettable synthetic dataset (36 reports, four illustrative sites).
The accessibility features are: Mobile layouts, keyboard navigation, dialog focus management, optional browser WebMCP tools.
The prototype's analysis engine is a deterministic English-language heuristic.
Its scores are not calibrated probabilities.
The prototype is not a trained NLP classifier or LLM verifier and it must not be used for operational safety decisions.
Rule tags denote matching activity context, not validated violations.
The graph groups shared tags rather than proving causation; similar reports use lexical overlap rather than embeddings.
The seven categories do not constitute complete IOGP rule coverage.
All application records are stored in the browser's local storage; they are not shared between devices or origins.
No production database, API server, user authentication, external alerts, active learning, multilingual model, or live HSSE integration is connected.
Original analysis is retained when a reviewer changes a label.
The dashboard uses the last 7 or 14 days; older imported reports are retained but may fall outside the visible period.
Synthetic sample dates span September 1–12, 2026.
I run the automated tests with node --test tests/*.test.mjs.
The tests cover source evidence, uncertainty, negation, recurrence windows, CSV structure and formula escaping.
The demo sequence is: Select New report, then Hot work + isolation and Analyze report; inspect the score, highlighted narrative, unverified isolation, and related reports; confirm the assessment or correct the SIF label with a reviewer note; open Recurrence radar, select a pattern, and inspect its connected sites; open Review & feedback to see the saved decision.
The source is modular: dist/engine.js contains analysis, dist/csv.js CSV parsing/export, and dist/app.js interface state and browser persistence. A future backend can replace these analysis and persistence boundaries with the PRD's API contract.
I use GitHub as Pinku Sharma (@Gradientblack9), a Computer Science student interested in coding, problem-solving, new technologies, software development, and AI. My profile lists eight repositories across JavaScript, HTML, Python, and TypeScript.
I describe myself as “a Computer Science student with a passion for coding and problem-solving.”
I love exploring new technologies, software development, and AI.
I have been on GitHub since 2024.
My GitHub profile has 1 follower.
I describe myself as a student web developer creating modern, brutalist-style websites with HTML and CSS.
The listed repository language counts are HTML (2), JavaScript (2), Python (1), and TypeScript (1).
My sentinel-prototype repository is JavaScript; last updated 2026-09.
My Pinku_Sharma_Portfolio repository is HTML; last updated 2026-09. Its description identifies me as a student web developer creating modern, brutalist-style websites with HTML and CSS.
My voice-rag-hacker-house-goa-2026 repository is Voice-enabled MSMARCO-XI RAG system; Python; last updated 2026-08.
I built this Python project as an evidence-first Voice RAG service for Hacker House Goa 2026. Its local-text benchmark, voice interface, ingestion path, and open validation work are described below.
My voice-rag-hacker-house-goa-2026 project architecture is an end-to-end, evidence-first Voice RAG service with replaceable Sarvam transcription, safety checks, hybrid retrieval, extractive/local generation, grounding verification, and typed API responses.
My voice-rag-hacker-house-goa-2026 default behavior is closed-corpus; answers questions present in the local MSMARCO-XI index and abstains when it cannot verify an answer.
My voice-rag-hacker-house-goa-2026 request path is voice → Sarvam STT → validation/guardrails → hybrid vector + keyword retrieval → grounded generator → grounding check → response.
My voice-rag-hacker-house-goa-2026 harness provides per-stage timing; provenance citations on every answer.
My voice-rag-hacker-house-goa-2026 assistant abstains for unsafe, off-topic, insufficient-evidence, or ungrounded requests.
My voice-rag-hacker-house-goa-2026 landing experience has selectable English, Hindi, Bengali, Tamil and Telugu speech/UI modes; Sarvam automatic language detection; persistent light/dark settings; microphone-reactive voice button glow follows live input volume.
My voice-rag-hacker-house-goa-2026 dataset ingestion uses the official ai4bharat/MSMARCO-XI dataset; prints available fields and indexes passage/text/context fields; offline indexing.
My voice-rag-hacker-house-goa-2026 chunking uses sentence-semantic chunks for short text; metadata-aware semantic chunks for structured records; overlapping sliding windows otherwise; chunks preserve source, document ID, strategy, and position.
My voice-rag-hacker-house-goa-2026 ingestion artifacts are data/index.json; data/index-lite.json; compressed generated Python module of verified answer rows for serverless deployment. The app reconstructs the embedded index when the full local index is unavailable.
My voice-rag-hacker-house-goa-2026 working examples are What is a corporation?; What is honesty?; Why did Rachel Carson write An Obligation to Endure?; Hey model greeting returns guidance directly.
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
pip install datasets
python scripts/ingest.py --limit 5000
uvicorn app.main:app --reload
My voice-rag-hacker-house-goa-2026 benchmark measured on August 19, 2026; P50 18.454 ms; P70 23.991 ms; P100/max 45.299 ms; mean 15.983 ms; errors 0 / 100. It used 100 distinct source queries across the 26,991-chunk index, disabled optional network fallback, excluded one-time index loading, and wrote data/benchmark_results.json.
My measured local result meets the <200 ms target for the local text retrieval, extractive generation, and grounding pipeline.
My benchmark is not presented as voice end-to-end; Sarvam remote STT must be benchmarked separately with real audio from final deployment region.
My next production step is to add a production LLM adapter behind app/generation/ for richer synthesis and re-run benchmarks.
I will do not claim voice target until real-audio benchmark verifies it; end-to-end under-200ms depends on Sarvam network latency and deployment region.
My default generation is intentionally extractive for fast, reproducible local operation. The query endpoint works without provider credentials, while voice queries use Sarvam and require SARVAM_API_KEY in a private .env. The metrics endpoint reports live P50/P70/P100. Open questions are real-audio results and end-to-end voice latency from the final deployment region, and the synthesis and benchmark results a production LLM adapter would provide.
My Hacker-House-2026 repository is Personalized ID card generator; HTML; last updated 2026-08.
The ID card generator flow is to upload a photo, add details, customize the card, preview it, download it, and share it on X.
My mini-content-engine repository is JavaScript; last updated 2026-07.
My cyber-sphere-enhanced-fixed repository is TypeScript; last updated 2026-04.
My Gradientblack9 repository contains GitHub profile configuration files; last updated 2025-03.
My Js---hindi repository is JavaScript code repository for the Chai and Code YouTube channel; last updated 2024-11.