Alchemyst Labs
Open-source applied AI research - done with taste.
Global research for the local scale
On-device models, local harnesses and everything in between.
pratilekha-v0
Whisper Tiny · 39M paramsLightweight Indic STT using LoRA adapters on Whisper Tiny
pratilekha-v0-small
Whisper Small · 244M paramsHigher-accuracy Indic STT using LoRA adapters on Whisper Small
bodh
Whisper Medium · ~1.5B paramsLightweight Indic language identity detection for VAD-to-STT routing. Frozen Whisper medium encoder with a learned projector and linear classifier — 98.81% accuracy on 7 Indic languages in 30-40ms
Alchemyst Labs - Research Wing of Alchemyst AI
Alchemyst Labs is the research division of Alchemyst AI (XAlchemyst Technologies Pvt. Ltd.), based in Bengaluru, India. We advance open-source AI research across Indic speech-to-text, context-aware distributed systems, and agentic AI infrastructure. Our motto is open models, open source, open science. Main site at labs.getalchemystai.com. Parent product at getalchemystai.com. Contact founders@getalchemystai.com for research collaborations, SDK support, and press inquiries about Alchemyst Labs.
Pratilekha Indic Speech-to-Text Models by Alchemyst Labs
Pratilekha, Sanskrit for transcription, is a family of lightweight Indic Speech-to-Text models by Alchemyst Labs, LoRA fine-tuned from OpenAI Whisper for edge and mobile inference. Variants include pratilekha-v0 built on Whisper Tiny with 39M parameters and pratilekha-v0-small built on Whisper Small with 244M parameters. They support 12 Indian languages: Hindi, English, Bengali, Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi, Urdu, and Odia. Benchmarks show around 20 percent better accuracy than Whisper-1 on Indic test sets with sub-200 millisecond latency on edge devices. Models are open on HuggingFace under anuran-roy namespace, with real-time WebSocket FastAPI server examples. Bodh is our lightweight Indic language identification model for VAD-to-STT routing, reaching 98.81 percent accuracy in 30 to 40 milliseconds on NVIDIA L40S hardware.
Open Source SDKs and Developer Resources from Alchemyst Labs
Alchemyst Labs maintains open-source developer tools for the Alchemyst Context Layer. TypeScript SDK at npmjs.com/package/@alchemystai/sdk provides context arithmetic and institutional knowledge graph access. Python SDK at pypi.org/project/alchemystai supports sync and async via httpx. Framework integrations include Vercel AI SDK middleware withAlchemyst and alchemystTools for streamText and generateText, plus LangChain JS AlchemystRetriever and AlchemystMemory for context-aware chains. GitHub organization github.com/Alchemyst-ai hosts 36 repositories including awesome-saas starter templates with 421 stars. MCP clients can discover native tools at labs.getalchemystai.com/.well-known/mcp via Streamable HTTP. Full docs index at labs.getalchemystai.com/llms.txt and detailed dump at labs.getalchemystai.com/llms-full.txt. Sitemap at labs.getalchemystai.com/sitemap.xml. Research papers at labs.getalchemystai.com/papers, models at labs.getalchemystai.com/models, open-source catalog at labs.getalchemystai.com/open-source, about at labs.getalchemystai.com/about, contact at labs.getalchemystai.com/contact, privacy at labs.getalchemystai.com/privacy. Frequently asked: what is Alchemyst Labs, what are Pratilekha models, which SDK languages are supported, and how to contact Alchemyst Labs for collaborations.