Harsh Joshi

    Engineering & dark arts · josharsh.com/ai

    Josh
    LM
    online · subsecond

    Powered by JoshLM-7B.

    A 7-billion-parameter language model fine-tuned on harsh.joshi.pth's life. State-of-the-art on the Joshi-Bench (n=1). Subsecond latency, zero hallucinations*, full transparency through the trace panel on the right.

    *Definitions of "language model" may vary. See the in the trace panel.

    Model loading status: idle

    Agent traceno actions yet
    Ask anything. Each query triggers bm25Search, optionally embeddingRerank, then composeResponse. Full timings and scores show here.

    JoshLM-7B · fine-tuned on a life of code · p95 < 400ms · 22 MB

    JoshLM-7B may be . For the source of truth, see the resume or email Harsh.