JobSignal — RAG sobre Ofertas de Emprego
chandrasekhar-jinendran/jobsignal
Chat com RAG sobre job postings, retrieval citado e stack análise
Ver no GitHubAplicação web que implementa um pipeline RAG completo (FastAPI + busca vetorial + Gemini/Claude) para analisar ofertas de emprego. Você cola descrições de vagas e faz perguntas cross-posting ("que skills aparecem em todas?", "sou candidato?", "qual stack querem?"), recebendo respostas citadas com fontes. Funciona em modo stub (sem API key) ou com modelos reais.
Instalação
npx convez-ai chandrasekhar-jinendran/jobsignal1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Analisar múltiplas ofertas de emprego e identificar skills comuns entre elas
- Avaliar fit pessoal versus requisitos de múltiplas vagas simultaneamente
- Extrair padrões de stack tecnológico em aberturas de um segmento
- Aprender padrão completo de RAG: chunking, embeddings, vector store, prompt engineering
Veredito

Projeto maduro e bem arquitetado que demonstra RAG em produção. Excelente referência prática para aprender o stack FastAPI+vector search+LLM, com fallback stub inteligente e suporte dual Gemini/Claude. API clara e UI funcional.
README
A tradução para o português deste README ainda está na fila.
JobSignal — chat over job postings with RAG
Paste job descriptions in, then ask questions across them — "which skills appear in every posting?", "am I a fit for these roles?", "what stack do they all want?". A retrieval-augmented-generation (RAG) pipeline finds the relevant excerpts and Claude answers from them, citing which posting each point came from.
Why this project: per the portfolio-project-finder market baseline
(../../references/market-signals.md), RAG + vector search + FastAPI is the
single most-demanded foundational skill cluster in current AI job postings —
and it was the biggest uncovered category in the portfolio
(../../references/project-log.md). JobSignal also dogfoods the portfolio
theme: it analyses the very job postings the skill scans.
Architecture
paste JD ─▶ POST /api/postings ─▶ chunker ─▶ embeddings ─▶ [vector store]
│ top-k
ask question ─▶ POST /api/chat ─▶ retrieve ─▶ prompt ─▶ Claude (or stub) ─▶ answer + sources
| Module | Role |
|---|---|
src/jobsignal/config.py | env-driven settings; no key → stub mode |
src/jobsignal/store.py | vector store: ChromaDB if installed, else built-in hashing-embedding store |
src/jobsignal/ingest.py | posting registry + paragraph-aware chunking |
src/jobsignal/rag.py | retrieve top-k chunks, hand to the LLM |
src/jobsignal/llm.py | Claude client + deterministic stub fallback |
src/jobsignal/app.py | FastAPI app + REST API |
src/jobsignal/static/index.html | single-page chat UI |
The LLM is stubbed by default: with no API key the whole pipeline runs and
the "answer" is a deterministic summary of what was retrieved. Provide a key
and nothing else changes. The provider is auto-detected — a Gemini key
(free tier via Google AI Studio) is used first, then an Anthropic/Claude
key, else stub. Force one with JOBSIGNAL_PROVIDER=gemini|claude|stub.
How to use it
1. Setup (Windows PowerShell)
cd Project\jobsignal
py -m venv .venv
.venv\Scripts\python.exe -m pip install -r requirements.txt
Optional — real vector DB and real answers:
.venv\Scripts\python.exe -m pip install chromadb # otherwise built-in store is used
copy .env.example .env # then put your ANTHROPIC_API_KEY in it
$env:ANTHROPIC_API_KEY = "sk-ant-..." # or set it for the session
2. Run
.venv\Scripts\python.exe -m uvicorn jobsignal.app:app --app-dir src --reload
Open http://127.0.0.1:8000 — left panel: paste postings; right panel: chat.
The header shows the current mode (stub or claude) and store backend.
3. Use the API directly (optional)
# add a posting
Invoke-RestMethod -Method Post -Uri http://127.0.0.1:8000/api/postings `
-ContentType 'application/json' `
-Body (@{title='GenAI Engineer'; company='Acme'; text='...full JD text...'} | ConvertTo-Json)
# ask a question
Invoke-RestMethod -Method Post -Uri http://127.0.0.1:8000/api/chat `
-ContentType 'application/json' `
-Body (@{question='Which skills recur across all postings?'} | ConvertTo-Json)
# health / list / delete
Invoke-RestMethod http://127.0.0.1:8000/api/health
Invoke-RestMethod http://127.0.0.1:8000/api/postings
Invoke-RestMethod -Method Delete http://127.0.0.1:8000/api/postings/<id>
4. Test
.venv\Scripts\python.exe -m pytest tests -q
Tests run fully offline (stub mode + built-in store).
Configuration
| Env var | Default | Meaning |
|---|---|---|
JOBSIGNAL_PROVIDER | auto | gemini, claude, or stub; auto = gemini key → claude key → stub |
GEMINI_API_KEY | — | free tier via Google AI Studio |
ANTHROPIC_API_KEY | — | optional Claude alternative (prepaid credit) |
JOBSIGNAL_MODEL | provider default | blank → gemini-2.5-flash / claude-opus-4-8. Gemini free: gemini-2.0-flash, gemini-1.5-flash. Claude cost-savers: claude-sonnet-4-6, claude-haiku-4-5 |
JOBSIGNAL_STORE | auto | chroma or simple |
JOBSIGNAL_DATA_DIR | ./data | where postings + vectors persist |
JOBSIGNAL_TOP_K | 6 | retrieved chunks per question |
JOBSIGNAL_MAX_TOKENS | 4096 | answer length cap |
Deploy story (free of cost)
Render free tier via the included render.yaml: New ▸ Blueprint ▸ pick
this repo. It sets build command pip install -r requirements.txt, start
command
uvicorn jobsignal.app:app --app-dir src --host 0.0.0.0 --port $PORT,
and JOBSIGNAL_PROVIDER=gemini. Add your GEMINI_API_KEY as a secret in the
dashboard (free tier — no card) and you have a live, fully-answering app at $0.
Note: the free tier has an ephemeral filesystem, so pasted postings reset on restart/redeploy — fine for a demo. Stretch goals for persistence: Supabase pgvector backend, plus a Dockerfile, basic auth, and a "skill-frequency dashboard" across stored postings.
Safety note
Job descriptions are public text, but don't paste content containing personal
data. Stored data lives unencrypted in JOBSIGNAL_DATA_DIR; treat a deployed
instance accordingly (add auth before sharing the URL).
Resume bullet (once deployed)
Built and deployed JobSignal, a RAG web app (FastAPI + vector search + Claude API) that ingests job postings and answers cross-posting skill queries with cited sources — live at , N postings indexed.