richartdo/ascent

Ascent is an AI-powered opportunity discovery and application assistant helping African students, graduates, and young founders find, match with, track, and apply for relevant opportunities. # Ascent Ascent helps African students, recent graduates, and young founders move from finding a verified opportunity to submitting a stronger application. Users create a profile, discover and save opportunities, compare profile fit, assess readiness, review a CV, and track every application through deadlines and checklist steps. **OpenAI Build Week track:** Education ## What works - Supabase email/password authentication and private user profiles - Verified opportunity discovery with search, filters, deadlines, and official application links - Saved opportunities with personal notes - Drag-and-drop application tracking across eight controlled statuses - Application checklists, dashboard reminders, and in-app notifications - Private PDF/DOCX CV upload with browser-side text extraction - Deterministic opportunity matching using a trained scikit-learn pipeline - Local opportunity summaries, readiness explanations, and CV analysis using Ollama Ascent does not submit applications on a user's behalf. It links to the official application page and helps the user prepare and track their progress. Cover-letter and essay generation are intentionally deferred rather than returning fake AI content. ## Built with Codex and GPT-5.6 Ascent was developed during OpenAI Build Week through an iterative Codex workflow using GPT-5.6. Codex accelerated repository analysis, API and schema implementation, Supabase migrations and RLS tests, frontend/backend integration, local-model evaluation, failure diagnosis, documentation, and automated verification. The builder retained the key product and engineering decisions: prioritize the profile-to-application journey, use only verified opportunities, keep user data protected by RLS, separate deterministic matching from generative analysis, run sensitive CV analysis locally, expose transparent readiness components, and disable features whose model quality was insufficient. GPT-5.6 is used through Codex as the engineering collaborator for …