Akshath Tiwari
← All projects

Production LLM Services

Production LLM FitScore Microservice

FastAPI service serving a fine-tuned LLM on vLLM for real-time job–candidate scoring, with structured output, skill-gap analysis, and automatic OpenAI fallback.

FastAPIPydantic v2vLLM (xgrammar)MongoDBRedisLangfuseOpenTelemetry

The production entry point for the FitScore system: a FastAPI microservice that extracts skills and entities from a job description and resume pair, performs multi-axis skill-gap analysis (exact / semantic / no-match), and produces a weighted, explainable fitscore with per-axis reasoning — all served by a fine-tuned model running on vLLM.

Includes a healthcare-specific scoring path and live prompt management via Langfuse’s prompt registry, so prompts can be iterated without a redeploy.

Key Results

  • →Skill-index token-efficiency encoding: model returns integer IDs instead of names, cutting output tokens substantially
  • →Structured JSON output via vLLM + xgrammar constrained decoding (Pydantic schema, strict mode)
  • →Automatic OpenAI fallback on token/context errors with a daily rate limit for safety
  • →Per-client configurable weighted scoring pulled live from MongoDB
  • →Parallelized entity/skill extraction via ThreadPoolExecutor for lower p95 latency