Your autonomous AI workforce that discovers opportunities, analyzes your profile, tailors your resume, and applies to jobs automatically while you focus on building your future.
Each agent owns a single stage and exposes a clean interface. The orchestrator coordinates them — it contains zero business logic itself.
Concurrently scrapes Remotive, Himalayas, Arbeitnow, and HN Who's Hiring. Queries the global job pool first — external APIs only fire when fewer than 5 local matches exist.
Scores every lead with a three-signal formula: 50% semantic similarity via Jina AI embeddings, 30% keyword overlap against your résumé skills, 20% title match. Assigns HOT / WARM / COLD / REJECT bands.
Fetches real-time DuckDuckGo headlines to overcome the LLM's knowledge cutoff, then calls Groq to produce a company stability score (0–100), tech stack, and funding timeline.
Runs a Groq → Gemini → HuggingFace waterfall with 3 retries per provider. HOT leads get a full résumé rewrite; WARM leads get a targeted skills highlight. All outputs are JSON-validated before saving.
Scores your tailored résumé against the job description (0–100 ATS compatibility), detects keyword gaps, and generates a role-specific interview prep cheat sheet with historical questions by stage.
Generates a pixel-perfect PDF via RenderCV (Typst backend) and uploads it to Supabase Storage. Sends interactive job cards to Telegram and fires cold emails via Gmail SMTP with the PDF attached.
Records every dismissal signal (too junior, wrong stack, bad company) and adjusts scoring weights per user — so the pipeline gets sharper with every interaction.
Aggregates HOT/WARM/COLD counts, status distributions, and score histograms via a high-performance Supabase RPC. Fires a daily digest to Telegram every morning at 09:00 IST.
Upload a PDF or paste your résumé JSON. Groq (Llama-3.1) parses it instantly into a structured profile. Your master embedding is computed once and cached in Supabase.
The global harvester runs across Remotive, Himalayas, Arbeitnow, and HN Who's Hiring in parallel. BM25 pre-filter drops irrelevant results before they ever touch the DB.
Jina AI embeddings compute cosine similarity between your résumé and every job description. Weighted 50% semantic + 30% keyword + 20% title match → final 0–100 score.
HOT leads (≥ your threshold + halfway to 100) get a full résumé rewrite. WARM leads get targeted highlights. All outputs pass a JSON structure validator before being saved.
RenderCV compiles your tailored résumé JSON into a Typst-rendered PDF. Uploaded to Supabase Storage with a permanent public URL. Theme is per-user configurable.
An ATS agent scores your tailored résumé (0–100), surfaces keyword gaps, and generates an interview prep playbook — cultural values, historical technical questions, recent launches.
Interactive job cards land in your Telegram DMs. Each card shows the score band, company, role, and a direct apply link. Notifications fire only above your custom threshold.
The pipeline hunts for a company email via OSINT, drafts a personalized follow-up via Groq, and dispatches via Gmail SMTP — PDF résumé attached.
PhantmOS monitors job markets across every major city and remote-first company, ensuring zero opportunity goes undetected.
A shared global_jobs table is populated once across all users. The per-user pipeline queries it locally first — external API calls only fire when fewer than 5 fresh matches exist. Zero redundant scraping.
Bring your own Groq, Gemini, or HuggingFace API keys. They are encrypted with Fernet (AES-128) before storage — a SHA-256 digest of your Supabase key is the symmetric seed. Only "***" is ever sent to your browser.
Tailored résumé JSON is compiled into a polished PDF via RenderCV's Typst backend. Theme is per-user configurable (sb2nov, classic, engineeringresumes). PDFs land in Supabase Storage with permanent public URLs.
HOT/WARM leads trigger Telegram job cards above your custom score threshold. The Phantm Writer endpoint drafts a personalized follow-up email, hunts the recruiter's address via OSINT, and dispatches via Gmail SMTP with the PDF attached.
I uploaded my PDF on Sunday. By Monday morning my Telegram had 12 job cards — each with a tailored résumé already generated. HOT leads had personalized cover emails queued. I hadn't touched a keyboard.
The scoring model is brutally honest. BM25 pre-filters noise before it reaches the DB, then Jina embeddings rank what's left by true semantic fit — not keyword stuffing. REJECTs never waste your time.
Groq fails? Gemini picks it up. Gemini rate-limits? HuggingFace steps in. Every output is JSON-validated before saving. One bad API call has never once stopped a pipeline run.
Upload your résumé. Set your target roles. PhantmOS runs the rest — discovery, scoring, tailoring, PDFs, and delivery — fully on autopilot.