Skip to content
View ramankrishna's full-sized avatar
7 followers · 12 following
  • United States
  • 17:29 (UTC -12:00)

Achievements

Achievement: Pair Extraordinairex2Achievement: Pull Sharkx2Achievement: YOLO

Achievements

Achievement: Pair Extraordinairex2Achievement: Pull Sharkx2Achievement: YOLO

Block or report ramankrishna

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ramankrishna/README.md

Rama Krishna Bachu — Training small specialized language models end-to-end

Software Engineer at Lowe's · MS CS, University of New Haven · Independent SLM research as Bottensor

Currently

Bottensor Open-source SLM research line. Training, serving, routing, and multi-agent orchestration as one stack. bottensor.xyz
polyrt Unified LLM runtime across MLX, Anthropic, and OpenAI. PyPI, v0.1.0.

Models on HuggingFace

Live download stats from the HuggingFace API, refreshed weekly.

NPC Fast 1.7B NPC Agentic 7B v3

Fin-PRM 7B NPC Fin 32B

Profile on huggingface.co/ramankrishna10

Published research

Date Title Summary Links
2026-04 NPC Fast 1.7B SmolLM2-1.7B-Instruct, full-weight CPT to 16K + router LoRA SFT. 98.3% OOD router accuracy, 77.8% IFEval. DOI · HF
2026-04 NPC Fin 32B Qwen2.5-32B-Instruct, 12×H100 ZeRO-3, 32K SFT examples. CryptoQA 93.6%, GPTQ 4-bit 19.2GB. DOI · HF
2026-04 Fin-PRM 7B Qwen2.5-7B QLoRA process reward model, 4-dim. Spearman 0.9234, F1 0.8421, ECE 0.21. DOI · HF
2026-05 NPC Agentic 7B v3 Qwen2.5-7B QLoRA + benchmark harness. BFCL v4 / IFEval / AgentBench under identical vLLM 0.6.3 GPTQ W4A16 serving. DOI · HF · bench

All preprints CC-BY-4.0 · Bottensor Independent Research · ORCID 0009-0000-1298-0681

Open source

Bottensor

bottensor-models npc-mom-router bottensor-fleet bench-npc-agentic-v3

@bottensor on npm

cogito engram forge

Standalone

polyRT

Experience

Role Where What
Software Engineer Lowe's, Charlotte NC Event-driven microservices, 1M+ Kafka events/day for promotions and payments. Postgres + Couchbase data sync. Grafana observability. 10M+ digital gift card batch processing.
Software Engineer Accenture, Hyderabad Data ingestion with Redis / Solr / Kafka — 140% query perf gain across 100M+ requests. AWS S3 + Avro pipelines. Fault-tolerant microservices serving 20M+ users.

Connect

Independent research. Affiliations represent personal interests, not employer.

Pinned Loading

  1. bottensor-models bottensor-models Public

    Training scripts and configs for the Bottensor NPC model family (npc-fast, npc-fin, npc-mom-router, npc-agentic v1/v2)

    Python

  2. coinbase/x402 coinbase/x402 Public

    Forked from x402-foundation/x402

    A payments protocol for the internet. Built on HTTP.

    TypeScript 87 101

  3. forge forge Public

    Open agentic framework — skills-as-markdown, pluggable LLM providers, file-based memory. @bottensor/forge on npm.

    TypeScript 1

  4. engram engram Public

    Graph-based agent memory — short-term and long-term memory with local or Voyage AI backends. @bottensor/engram on npm.

    TypeScript 1 2

  5. cogito cogito Public

    Multi-strategy reasoning for AI agents — ReAct, Chain-of-Thought, Tree-of-Thought with HuggingFace and OpenAI-compatible backends. @bottensor/cogito on npm.

    TypeScript