AI engineer / builder / product-minded developer

Bhavesh Goel

I build agentic systems, applied ML tools, and product-focused software that is meant to be used, not just demoed.

See projectsGet in touch

What I focus on

Areas of depth

🔗

Agentic systems

Workflows with clear orchestration, persistence, and delivery paths instead of one-off demos.

🧠

Applied ML

Projects that use models in real product loops, from STEM animation generation to computer-vision control.

🚀

Product-minded delivery

Interfaces, APIs, and operational details designed so the build can be shown, tested, and extended.

Selected work

Projects that stand on their own

Real builds with enough detail for a recruiter, founder, or hiring manager to actually dig into.

All projects
AI video pipelineFeatured
Updated Dec 2025

Live demo

NeuroAnim

Turns a topic prompt into an educational animation with narration, quiz questions, and final video output.

Built a Gradio-based generation system that plans the lesson, writes Manim code, renders video, synthesizes audio, and packages the result into a usable STEM explainer.

  • Prompt to concept plan to narration to Manim code to render to voiceover to quiz flow is wired end to end.
  • Includes audience and quality controls plus progress-aware status updates in the Gradio interface.
  • Supports cloud rendering and text-to-speech integrations for a production-style content pipeline.
PythonManimLangGraphGradioElevenLabsBlaxel
Live projectSource code
Agentic workflowFeatured
Updated Mar 2026

Active build

Naradbot

Personal agent that scores Hacker News, summarizes signal, and ships digests into Discord.

Designed a LangGraph-driven workflow with FastAPI, Discord delivery, scheduler support, scoring branches, and persistence so the system can run as an actual assistant instead of a toy chat loop.

  • Readable graph nodes handle fetch, dedupe, scoring, categorization, summarization, publishing, and run persistence.
  • Parallel editorial and opportunity scoring branches merge before digest generation.
  • Supports scheduled jobs, manual API triggers, Discord posting, and deployment paths for always-on or Blaxel job execution.
PythonFastAPILangGraphDiscord.pySQLiteAPScheduler
Source code

More work

Supporting builds

Streaming system
Updated Mar 2026

Public repo

P2P Streamer

Torrent-backed video player that starts fast, supports subtitles, and falls back to live transcoding.

Built a Node server plus React client that ingests magnet links or .torrent files, prioritizes playback-friendly pieces, and streams media with range support before the full download completes.

  • Accepts magnet URIs and .torrent uploads, then persists sessions and watch progress in SQLite.
  • Direct play uses HTTP range responses; unsupported formats fall back to on-the-fly ffmpeg transcoding.
  • Sidecar subtitles are exposed as WebVTT for browser playback and external players.
Node.jsReactViteWebTorrentffmpegSQLite
Source code
Computer vision utility
Updated Mar 2026

Repo coming soon

Gesture Cursor

Hand-tracked cursor control with gesture-based clicks and runtime tuning for real desktop use.

Built a low-latency computer vision loop that maps finger motion to cursor movement, toggles tracking with a held pinch, and supports click gestures without requiring a mouse.

  • Threaded camera capture keeps latency low while the hand-tracking loop stays responsive.
  • Supports both MediaPipe legacy solutions and Tasks API builds, including first-run model download.
  • Handles left-click, right-click, smoothing, cursor margins, and backend selection for X11, Wayland, or macOS.
PythonOpenCVMediaPipeComputer VisionAutomation
Repo coming soon