Home / About
About TabotecAn AI startup that ships, not slides.
Tabotec builds AI products and partners with teams adopting AI. Our products are production software you can run on your own hardware. Our consulting work is hands-on engineering, not strategy decks.
What we are
Two arms, one engineering bench.
Tabotec is structured around two arms. The products arm ships software — Tabotec Atlas (enterprise RAG), the Tabotec Languages program, and bespoke fine-tunes. The consulting arm rents that same engineering team to companies putting AI into their own products.
The two arms feed each other. Building our own products keeps our engineers sharp on production AI. Consulting engagements expose us to a wider variety of problems and feed lessons back into the product line.
Approach
Boring infrastructure. Honest evals.
We don't sell magic. We curate data carefully, pick well-understood base models, run real evals, and ship things that pass them. Every model has a confidence policy. Every data point has provenance. Every claim is testable.
Where it's defensible to be open, we are: open weights, public benchmarks, contributed datasets where licensing allows. Where it isn't — your private corpus, your customer recordings — we run on your hardware and never see it.
Where we are
Headquartered in Los Angeles.
Tabotec is headquartered in Los Angeles, California, with a language-focused subsidiary in Addis Ababa, Ethiopia. The geographic spread shapes who we partner with regionally — EAII, Lesan AI, Masakhane on the language side — but our products and consulting practice serve teams worldwide; we're not a regional shop.
Product roadmap
Tabotec Atlas. Enterprise RAG, shipping today, evolving toward agent workflows and richer multi-source ingestion.
Tabotec Languages — Phase 1. Tigrigna speech (Tigvoice ASR + TTS) and language (Tabotec Tigrigna) in pilot 2026.
Tabotec Languages — Phase 2. Amharic, Afaan Oromoo, Somali. Same pipeline.
Tabotec Languages — Phase 3. Long tail — Afar, Sidamo, Wolaytta, Hadiyya, Sidaama. Cross-pollinate with regional partners.
Custom fine-tuning, always. The same pipeline applied to any team's data and domain.
Consulting practice
Most engagements run 4–12 weeks. We do discovery sprints, build engagements, and fractional-AI-lead retainers. Vendor-neutral on models and infrastructure.
Partners we're talking with
- Ethiopian Artificial Intelligence Institute (EAII)
- Lesan AI — machine translation, DAIR partner
- Masakhane — African languages research collective
- SERET Foundation — Tigrigna NLP community
- Regional broadcasters and ministries
What we believe
Three opinions that shape what we ship.
Small, owned models beat big rented ones.
For most production use cases, a 4–8B model fine-tuned on the right data outperforms a frontier API by every metric that matters at scale: latency, cost, privacy, and accountability.
Data is the product.
The model is downstream of the data. We invest most of our energy in curation, alignment, and review — because that's where quality actually comes from.
Local-first, always.
Most of our customers can't ship their data to a US cloud. We design for that constraint and treat it as a feature, not a limitation.
Want to work with us?
We're hiring researchers and engineers, looking for design partners for our product line, and open to consulting engagements. Drop us a line.