Skip to content
View vijaxx's full-sized avatar

Block or report vijaxx

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
vijaxx/README.md

Hi, I'm Vijay 👋

Computer Science Engineering graduate (B.Tech, July 2026 — GITAM Deemed University, Hyderabad), based in India and open to relocation.

I build systems that run themselves — pipelines that go from raw input to a finished, published product with no manual step in between — and I write my READMEs the way I'd want a reviewer to read code: what actually runs, what's measured, and what isn't there yet.

Open to: data & analytics, backend, full-stack, and AI-automation roles.


Live in production

Multi-Platform Content Automation Pipeline Sources stock footage, renders a captioned 9:16 video with FFmpeg, writes platform-tailored copy with Claude, and publishes to YouTube Shorts, Rumble, and Facebook Reels on a cron schedule. YouTube goes through the Data API; Rumble and Facebook ship no usable upload API, so those are driven over the Chrome DevTools Protocol against a real logged-in browser — the only approach that survives Cloudflare bot detection and React's rejection of synthetic clicks.

redditreels Finds a story, rewrites it into 45-second hook-first narration with an LLM, generates voiceover with per-word timing, renders karaoke-style animated captions, publishes across platforms — then measures how each upload performed and feeds that back into next week's source, title, and hashtag choices. Includes a self-healing layer that pauses the pipeline automatically after a failure streak rather than publishing garbage on a schedule.

pinforge Turns one theme into a complete digital product line: print-ready PDF, Pinterest pin creatives, SEO copy, and a manifest — then posts the pins and tracks performance in SQLite (impressions, saves, clicks per pin; sales by referrer).

kdp-puzzle-engine Generates print-ready puzzle books (word search, Sudoku, mazes) end to end at 300 DPI, with an unattended factory mode that rotates recipes to avoid near-duplicate titles, and a pre-upload compliance check that catches the specific formatting issues that trigger Amazon KDP review flags.

stylebyclaud A zero-dependency affiliate storefront, machine-generated and deployed on GitHub Pages. Live: vijaxx.github.io/stylebyclaud

Three of these repos carry green CI, unit tests, and an MIT license. Recurring pattern across all five: Chrome DevTools Protocol automation for platforms with no workable API, a provider-agnostic LLM shim (one interface, routes to whichever of Anthropic/Gemini/Groq/Ollama has a configured key), and failing safe — dry-run modes, pre-publish sanity checks, automatic pause on failure streaks.


Applied projects

Built to work end-to-end, with real test suites and measured results rather than illustrative numbers — each README states plainly what was run and verified versus what wasn't.

book-demand-predictor — Flask app pairing content-based recommendations (TF-IDF + cosine similarity) with monthly demand forecasting (Ridge regression on lag and seasonality features). Validated on a temporal holdout against a naive last-month baseline — R² 0.883, 14% better MAE than the baseline. Chart.js admin dashboard, 21 tests including target-leakage guards.

algo-trading-system — backtesting engine for five intraday strategies (ORB, EMA+RSI, VWAP reversion, Supertrend, Bollinger) against a full Indian transaction-cost model (STT, GST, stamp duty, brokerage, exchange charges). 92 tests, including a no-lookahead proof and a real concurrency bug caught by the test suite and fixed (a stale position-count race that could oversubscribe the risk cap). Results are reported honestly — two of five strategies lose money after real costs, which is in the README rather than filtered out. Paper trading only; no live order routing.

supermart-erp — Java Servlets/JSP/JDBC MVC app: employee and inventory management, session-based auth with role separation, JOIN-based reporting. 60 tests plus live curl verification against the running server, including a SQL-injection payload attempted against the real login endpoint and confirmed rejected.

library-management-system — Java Swing + JDBC desktop app: transactional book issue/return workflow, per-tier borrowing limits, overdue fine calculation. 44 tests, including a rollback test built around a JDBC proxy that fails mid-transaction to prove a partial issue never persists.

reels-scheduler — Instagram Reels scheduling and publishing pipeline: SQLite-backed content queue, timezone-aware slot scheduling, content-hash duplicate prevention, dry-run-by-default publisher. 80 tests, including a full clone-and-rerun from a bare git clone with zero config to confirm it's genuinely reproducible, not just working in one workspace.

marketgen — client-onboarding platform: a 7-step, 87-question wizard with conditional branching and resumable sessions, multi-tenant data isolation, and a content-generation pipeline behind a provider interface (deterministic offline stub by default). 63 tests plus a live browser walkthrough of the wizard's branching logic. Real Postgres row-level-security policies are included as reviewed SQL but not executed (no live Postgres in this environment) — stated directly in the README rather than implied as tested.

wsn-secure-routing — WSN routing simulation comparing baseline LEACH against a hardened variant (RSA-1024 mutual auth, AES-GCM, an Interlock Protocol implementation resisting man-in-the-middle key exchange). 71 tests including live MITM-detection scenarios. Measured result: the secure variant costs ~5.5% more energy under normal conditions, but is ~8% more energy-efficient than baseline under an active sybil attack, since it rejects 100% of forged node identities that the baseline wastes energy servicing. Reported as measured, not tuned to a target figure.


Tech

Python · Java · SQL · SQLite / H2 / PostgreSQL · HTML/CSS/JS

Flask · FastAPI · Servlets/JSP · Swing — application frameworks Pandas · NumPy · scikit-learn · Chart.js — data analysis, ML, reporting FFmpeg · Pillow · reportlab — media & document generation Chrome DevTools Protocol · Selenium — browser automation LLM APIs — Anthropic, Gemini, Groq, local Ollama JUnit 5 · pytest — every repo above ships with a real, passing test suite git · GitHub Actions


📫 LinkedIn · Hyderabad, India · willing to relocate

Pinned Loading

  1. kdp-puzzle-engine kdp-puzzle-engine Public

    Generates print-ready large-print puzzle books (word search, Sudoku, mazes) for Amazon KDP, with unattended factory-mode automation and pre-upload compliance checks.

    Python 5

  2. pinforge pinforge Public

    Pinterest-driven pipeline for print-on-demand puzzle products: generates puzzle-pack PDFs, Pinterest pin creatives, and SEO copy, then posts and tracks performance.

    Python 2

  3. redditreels redditreels Public

    Unattended pipeline: sources a story, rewrites it with an LLM, generates voiceover, renders captioned vertical video, publishes to YouTube/Facebook/Rumble, and learns from performance data.

    Python 1