AI adoption & team enablement
I work inside companies — discovery, architecture, delivery, handover — and teach engineering and business teams to build, run, and own AI.
Founder, Vector Rubric
Founding Head of Platform & Client
Engineering, Maio Labs
I help companies own their AI.
I build agentic systems, voice and document intelligence, and the memory and search layers underneath — then teach teams to run them. Based in Portugal, working across Europe.
65%Lower pagination costs at NWZ
1,800Print pages automated per month
100M+Tokens a month on owned hardware
0Cloud invoice at Maio Labs
01 / Capabilities
From strategy to production.
One person across the model, the data, the agents, the interface, and the team that runs it. I build the system and teach the people who own it.
Agentic systems
Multi-agent workflows, tool use, MCP, and agents embedded in company-wide systems and intelligence.
Voice AI
STT, TTS and full-duplex voice with barge-in and live transcripts. Local ASR or cloud providers like ElevenLabs.
Document intelligence
OCR, VLM and LLM pipelines for structured extraction, digitisation and parallel processing with RabbitMQ.
Memory, knowledge & search
Knowledge bases with Cognee, agent memory with Hindsight, semantic and hybrid search with OpenSearch.
AI infrastructure & full-stack
Private cloud, bare-metal GPU clusters, vLLM, DGX Spark, local inference, observability, and the product on top.
02 / Selected work
Proof, not promises.
Systems in production, measured by what changed: cost, throughput, deadlines, and the teams that kept running them.
Eidosmedia2023 — 2026
Automatic Pagination
Built the system that integrated Sophi.io's AI layout engine into Eidosmedia Méthode. In production at Nordwest-Zeitung, it lays out ~1,800 pages a month, cuts pagination costs by 65%, and leaves editors full control of the page. The approach has since been adopted by news groups across Germany and France.
65%Lower pagination costs
1,800Pages per month
Maio Labs2026 — Present
Platform & Client Engineering
Founding Head of Platform & Client Engineering. I own the teams that ship AI into customers' businesses and the platform it runs on: private cloud, bare-metal GPU clusters, local model hosting, data pipelines, and production AI across industrial, automotive, food & beverage, healthcare, and professional services. Hundreds of millions of tokens a month on hardware in the room. Cloud invoice: zero.
100M+Tokens per month
0Cloud invoice
Vector Rubric2026 — Present
Founder
Knowledge bases, memory layers, agentic systems, voice and AI infrastructure for companies that want to own their intelligence instead of renting it. I build the system, then teach the team to run it.
Open source & communityOngoing
Build in public
Contributor to OpenSearch, Apache Commons, Jackson and Lecta. Published a welding-defect object-detection dataset on Hugging Face. Built Hermes Live Desktop for full-duplex voice. I write and benchmark local inference on DGX Spark.
03 / Teaching
I don't just build.
I don't just build.
I teach.
The gap isn't technology — it's fluency. I bring AI into companies by training the teams that use it, from the C-suite to the codebase.
Engineer training
Inference deployment, evaluation, prompt engineering, agents. Hands-on with real models.
Cross-functional
Product, ops, sales, leadership — everyone learns the AI layer of their role.
Workshops & labs
Immersive sessions. Build real AI features, not slides.
Embedded mentoring
I work alongside your team until AI becomes muscle memory.
04 / Toolbox
The stack I reach for.
I pick the smallest stack that survives production. These are the tools I use most, from the model layer to the interface.
Agents & orchestration
Memory & retrieval
Voice & realtime
Documents & data
Infrastructure
Engineering
05 / Writing & video
I think in public.
Technical deep dives, benchmarks, and honest takes on where agents, memory and AI infrastructure are actually going.
Your AI Agent Needs a Brain That Never Sleeps
Why agents need one memory that follows them across devices, and how I run mine on Hindsight with 50,000 memories.
Git Can't Handle AI Agents. Can DeltaDB or Entire?
DeltaDB vs Entire: agent state, memory, review, hosting, and the catch. What version control looks like when agents write the code.
Selected writingrikkarth.com/blog →
Benchmark results for Qwen3.6-35B-A3B-FP8 on DGX Spark via vLLMInference
Real Agency Is a Loop, Not a PromptAgents
Benchmark results for AEON-7 Qwen3.6-35B-A3B heretic NVFP4 + DFlash on
DGX SparkInference
How Lecta built its runtime MDX rendererEngineering
Your Company Isn't Slow. It's Scared.Strategy
06 / Timeline
The path here.
A non-linear route through hardware, software, and AI — which is why the stack never scares me.
2026 — Present
Founder, Vector Rubric
Knowledge bases, memory layers, agentic systems, voice and AI infrastructure for companies that want to own their intelligence.
2026 — Present
Founding Head of Platform & Client Engineering, Maio Labs
Client Engineering, Vision Architects, private cloud, bare-metal GPU clusters, local AI stack, and production AI. Also Founding Software & AI Engineer since March 2026.
2026
Software & AI Engineer, Orbit Platform
Short engineering stint on the team behind Lecta.
2023 — 2026
Eidosmedia — Automatic Pagination & AI Engineering
Built Automatic Pagination end to end and integrated Sophi.io into Méthode. Later, AI Software Engineer in R&D: semantic and hybrid search with OpenSearch, MCP, LLMs, autonomous agents and chatbots.
2022
Full-stack bootcamp, Code for All
Formal software engineering after a decade of electronics and hands-on technical work.
2009 — 2022
Before software
Six years in the Portuguese Army, then electronics, camera repair and technical support. The hardware years are why I'm comfortable with bare metal, GPUs and racks.