TextSense — NLP Microservice

TextSense is what an ML service looks like when a backend engineer builds it: the model is the easy part, and everything around it — health probes, metrics, graceful degradation, payload hardening — is the actual engineering.

The fallback is the feature

Most NLP "APIs" die the day the model download fails. TextSense wraps inference in a strategy pattern: it tries the transformers backend (DistilBERT SST-2 on CPU) and, if torch is unavailable or inference throws, degrades to a deterministic lexicon engine written in pure stdlib. The HTTP contract never changes — every response carries a model id (distilbert-sst2 or lexicon-v1) so callers know who answered and can decide what the reduced confidence means for them.

The lexicon engine is genuinely useful too: negation handling, intensifiers, punctuation emphasis — deterministic enough that the entire test suite runs offline in under a second, no gigabyte downloads, no GPU.

What the service ships

• POST /api/v1/sentiment — single text → label, confidence, latency, request id
• POST /api/v1/sentiment/batch — up to 64 texts, order preserved, cache-aware
• GET /metrics — Prometheus format: request counters, latency, cache hits
• GET /health — backend id + uptime for orchestrators
• Optional X-API-Key auth, payload caps with honest 413s

Why this belongs in my portfolio

Because "AI Specialist" claims are cheap and running services are not. TextSense shows the full path: model choice, fallback strategy, batch coalescing, response caching, metrics exposition and a test suite that treats determinism as a feature. It pairs with TaskFlow API as the ML sibling of my backend stack.

Try it

docker compose up --build, then post a sentence to /api/v1/sentiment — github.com/Godde3s/textsense