Ship fast. Keep it simple. Open source everything.
┌──────────────────────────────────────────────────────────────────┐
│ $ ssh git@github.com -p 443 │
│ PTY allocation request failed on channel 0 │
│ Hi AmitoVrito! You've successfully authenticated, but │
│ GitHub does not provide shell access. │
│ │
│ ...unless you know where to look. │
└──────────────────────────────────────────────────────────────────┘
nautiverse:~$ whoami+ AmitoVrito — Senior AI Specialist · Technical PM · Founder
Building production infrastructure for LLM systems — and open sourcing it.nautiverse:~$ systemctl status synapsekit --no-pager+ ● synapsekit.service — async-native Python LLM framework+ Loaded: loaded (/usr/local/lib/python3.12/site-packages/synapsekit)+ Active: ● active (running) since 2024-01-01; 2y 3mo ago
Process: ExecStartPre=uv run pytest tests/ -q (code=exited, status=0/OK)
+ Main PID: 1 (python)++ providers 27 openai · anthropic · ollama · gemini · cohere · mistral+ bedrock · azure · groq · deepseek · openrouter · together+ fireworks · cerebras · cloudflare · moonshot · perplexity+ vertexai · zhipu · ai21 · databricks · baidu · llamacpp+ minimax · alephalpha · huggingface · sambanova+ tools 41 calculator · python_repl · web_search · sql · http · shell+ arxiv · pubmed · wolfram · wikipedia · tavily · brave ...+ loaders 18 pdf · html · csv · json · yaml · xml · discord · gdrive+ audio · video · excel · powerpoint · docx · web · dir ...+ v.stores 9 inmemory · chroma · faiss · qdrant · pinecone · weaviate+ pgvector · milvus · lancedb+ hard deps 2 numpy · rank-bm25+ test suite ✓ 1450 passing
nautiverse:~$ systemctl status chunkrank --no-pager+ ● chunkrank.service — model-aware chunking and answer re-ranking for LLM pipelines+ Loaded: loaded (/usr/local/lib/python3.12/site-packages/chunkrank)+ Active: ● active (running)+ Main PID: 2 (python)++ purpose automatically adapts chunk size to the model's tokenizer and+ context window, then consolidates and ranks answers across chunks+ version v1.0.0
nautiverse:~$ systemctl status traceprop --no-pager+ ● traceprop.service — computation-level data lineage, gradient attribution and unlearning+ Loaded: loaded (/usr/local/lib/python3.12/site-packages/traceprop)+ Active: ● active (running)+ Main PID: 3 (python)++ attribution LDS 0.976 (Covertype 50K) · 0.884 (Adult Income) · CPU-only, no GPU+ unlearning >100% gap-closed on real data · provenance-guided gradient correction+ compliance EU AI Act Article 26 audit trail · structured JSON export+ backends numpy · pytorch · jax+ paper submitted · Proceedings of the VLDB Endowment Vol. 20 (VLDB 2027)+ version v0.6.0
nautiverse:~$ cat Dockerfile
FROM python:3.12-slim
LABEL maintainer="AmitoVrito"LABEL version="2026.1"LABEL description="Senior AI Specialist · Technical PM · Founder"RUN apt-get install -y \
critical-thinking \
async-execution \
open-source-mindset \
production-focus \
zero-magic
COPY ./experience /prod/background/
COPY ./synapsekit /prod/flagship/
COPY ./chunkrank /prod/tools/
COPY ./traceprop /prod/research/
ENV MODE=production \
SHIPPING=true \
MAGIC=false \
LOCK_IN=false \
FIRST_PRINCIPLES=true
EXPOSE 443/ideas
EXPOSE 8080/collaboration
HEALTHCHECK --interval=30d CMD uv run pytest tests/ -q || exit 1
CMD ["python", "-m", "ship_features", "--no-magic"]nautiverse:~$ cat /proc/self/focus
+ LLM Infrastructure → RAG · streaming agents · graph workflows+ ChunkRank → model-aware chunking · answer re-ranking across chunks+ Traceprop → data lineage · gradient attribution · unlearning · VLDB 2027+ Cost intelligence → CostRouter · BudgetGuard · FallbackChain+ EU Compliance → GDPR toolkit · EU AI Act risk classifier+ Eval-driven dev → EvalCI · regression gates · synapsekit test+ Product → AI strategy for regulated industriesnautiverse:~$ cat stack.json
{
"languages": ["Python", "TypeScript", "Scala", "Java"],
"ai_ml": ["PyTorch", "HuggingFace", "OpenAI", "Anthropic"],
"infrastructure": ["AWS", "Azure", "Docker", "Kubernetes", "Terraform"],
"data": ["PostgreSQL", "Redis", "Elasticsearch", "Kafka", "Flink"]
}nautiverse:~$ crontab -l
# nautiverse crontab — m h dom mon dow command
0 6 ** 1-5 /usr/bin/python review_prs.py --notify
30 9 *** /usr/local/bin/ship_feature --increment minor
0 10 *** /opt/synapsekit/bin/uv run pytest tests/ -q
@daily /usr/bin/think --deep --no-interrupts --duration 2h
@weekly /usr/local/bin/cut_release --tag v$(date +%Y.%W)
0 20 ** 5 /usr/local/bin/write_changelog --verbosenautiverse:~$ env | grep -v SECRET
SHELL=/bin/zsh
EDITOR=vim
LANG=python3.12
PATH=/prod:/oss:/research
MODE=production
SHIPPING=always
MAGIC=false
LOCK_IN=false
FIRST_PRINCIPLES=true
ASYNC=true
STREAMING=first
HARD_DEPS=2
INTERRUPTS=0
CAFFEINE=requirednautiverse:~$ git log --oneline --graph --all
* d8301c6 (HEAD -> main, tag: v1.4.6) SynapseKit v1.4.6 — subgraph error handling
* e1eb0df (tag: v1.4.5) SynapseKit v1.4.5 — 9 vector stores
* 7477bd9 (tag: v1.4.4) SynapseKit v1.4.4 — SambaNova, GoogleDrive
* a936078 (tag: v1.4.2) SynapseKit v1.4.2 — 27 providers · 41 tools · 1450 tests
* 3c4d5e6 (tag: v1.3.0) SynapseKit v1.3.0 — cost routing · eval CLI · MCP · A2A
* 1a2b3c4 (tag: v2024.x) AI infrastructure at scale — fintech & healthcare
* c8f3a91 (tag: v1.0.0) ChunkRank v1.0.0 — model-aware chunking and answer re-ranking
* 4d17279 (tag: v0.6.0) Traceprop v0.6.0 — lineage · attribution · unlearning · VLDB 2027nautiverse:~$ kubectl describe pod/amitovrito
Name: amitovritoNamespace: productionLabels: role: ai-specialisttype: founderstack: pythonAnnotations: ships: fastmagic: "false"Status: RunningContainers:
amitovrito:
Image: amitovrito:v2026.1Port: 443/TCPLimits:
cpu: unlimited # self-managedmemory: 16Gi # coffee-backedRequests:
cpu: 1 deep-focus-blockmemory: uninterrupted-timeReadinessProbe: http-get /health/synapsekit delay=0s period=30dLivenessProbe: exec [uv, run, pytest, tests/, -q] period=7dVolumes:
/prod/synapsekit ReadWriteMany/prod/chunkrank ReadWriteMany/prod/traceprop ReadWriteMany/etc/experience ReadOnlyEvents:
2024-01-01 Normal Started synapsekit — first commit2025-01-01 Normal Upgraded synapsekit v1.3 — cost routing, eval CLI, MCP2026-03-31 Normal Upgraded synapsekit v1.4.6 — 27 providers, 9 vector stores, 1450 tests2026-04-01 Normal Published chunkrank v1.0.0 — model-aware chunking and answer re-ranking2026-05-20 Normal Published traceprop v0.6.0 — data lineage, attribution, unlearning2026-Q2 Normal Building synapsekit v1.5.0 — EU compliance platformnautiverse:~$ man amitovrito
AMITOVRITO(1) User Commands AMITOVRITO(1)
NAME
amitovrito — senior AI specialist, technical PM, founder
SYNOPSIS
amitovrito [--oss] [--consulting] [--collaborate] [--hire]
[--no-magic] [--async] [--stream]
DESCRIPTION
Builds production infrastructure for LLM systems and open
sources it. Maintainer of SynapseKit. Author of ChunkRank.
Author of Traceprop. Operates in production mode. No magic.
No lock-in. Everything composable.
OPTIONS
--oss contribute to or use SynapseKit / ChunkRank / Traceprop
--consulting AI strategy for regulated industries
--collaborate joint research or product work
--hire see /proc/self/focus for current bandwidth
--no-magic default. transparent, plain Python only
--async all I/O is non-blocking
--stream token-level output by default
ENVIRONMENT
MODE=production SHIPPING=true MAGIC=false
FILES
/prod/synapsekit async-native Python LLM framework
/prod/chunkrank model-aware chunking and answer re-ranking
/prod/traceprop data lineage, gradient attribution, unlearning
/etc/experience read-only
BUGS
Occasionally ships too fast. Known issue. Won't fix.
SEE ALSO
synapsekit(3), chunkrank(3), traceprop(3), pypi(1)
AMITOVRITO(1) v2026.1 AMITOVRITO(1)
nautiverse:~$ █