Advanced Nonlinear Technologies
The model layer,
reinvented.
The Fern series of nonlinear models, perfected for edge deployment and better inference.
The shift
AI is leaving
the data center.
80%
Inference, not training
The operating cost of AI — and where almost all the spend goes.
70%
Of pilots fail
Cost. Privacy. Latency. The three forces collapsing the cloud-AI default.
3 forces
Cost · privacy · latency
Each one pushes AI off the cloud and onto the customer's environment.
The opportunity
Our slice
by 2030.
$60–119B
Edge AI market size by 2030
~20% CAGR
Fastest segment in AI
$50B+
On-prem hybrid AI in 2026
$300B
Sovereign AI infra by 2040
The position
Below
the floor.
Phi · Mistral · Gemma · Llama1B – 14B params
GPT-4 · Claude · DeepSeek100B – 1T+
1M
100M
10B
1T
Everyone else is competing in the 1–10B band. Fern targets 10M–100M, two orders of magnitude smaller. Same frontier capability, edge-deployable scale.
The thesis
Models,
not APIs.
Cloud AI ships a meter.
Nonlinear AI ships a model.
Frontier capability → ships to the customer's environment.
Their data → never leaves it.
The model → improves with use.
The brand architecture
One model family.
Two products.
Fern
The nonlinear model series
Fernfly
For developers
Ultra-low-parameter Fern models built and deployed by developers. Drop-in API. Free per call.
EverydaySeries
For workplaces
Agentic platform for the workplace. Builds a per-org Fern series on the customer's own data. Runs agentic workflows — summaries, approvals, automation — inside their stack.
Go-to-market · Fernfly
Drop in.
Ship in.
Wedge
Free signup. One API key. Developers drop in an ultra-low-parameter Fern.
Mechanic
Free per call, sustainable because the model is tiny. Apps ship intent-to-action AI in an afternoon.
Outcome
Spreads inside engineering teams the way Stripe and Vercel did.
In pilot: a Fixed Income Trading Desk in New York. 12M Fern-FC handling repeat queries against RiskVal + Bloomberg. Local. Per-call cost zero.
fernfly.com
Go-to-market · EverydaySeries
Agentic AI
for the workplace.
Wedge
EverydaySeries deploys inside the operator's stack. Day-one cost reduction on agentic LLM usage.
Mechanic
Builds a per-org Fern series tuned on the customer's own data. The platform learns the workplace.
Outcome
Custom Fern that stays inside the company. Agentic workflows. Co-ownership, not reselling.
What it does inside the workplace: daily standup summaries · PO approvals · QA across teams · per-employee agents · custom workflows that learn from corrections.
10M
Images in one per-org Fern
Live customer: a hotel food-waste analytics company in Dubai. 50MB on-device Fern-Vision trained on the customer's 10M annotated images. Works without wifi.
everydayseries.com
The flywheel
Every customer makes
every Fern smarter.
01
Customer runs any LLM
EverydaySeries runs frontier APIs, open-weights, or Fern. On their device or local hardware.
→
02
Wrong answers surface
Misclassifications, uncertain calls, edge cases. Captured at the source.
→
03
EverydaySeries retrains
Real production errors become labelled training samples. Per-org Fern trains on what matters.
→
04
Smarter Fern ships back
Customer's device or app updates. Cost per correct call drops. The loop starts again.
The architecture is the entry moat. The flywheel is the compounding moat.
Receipts
Lab to live.
In the lab — methodology benchmarks
Text encoder · BGE-large
7.2×
99.5% quality retained
Decoder LM · Pythia
11.4×
~90% retained
Vision encoder · DINOv2
9.4×
80% top-5 retained
In production — shipped customer models
Fern-Vision · food-waste
50MB
on-device, no-wifi kitchen
Fern-FC · trading desk
12M
local, repeat queries, $0/call
Delivery time
Months
custom Fern, not years
Competition
Different market.
Different economics.
| OpenAI / Anthropic | DeepSeek / Kimi | Thinking Machines | ANT |
|---|
| Model size at deployment | Huge | Large (up to 685B) | Same as base | Smaller than the base |
| Platform around the model | API only | None | API only | EverydaySeries + Fernfly |
| How you use it | API only | DIY weights | API for engineers | DIY or done-for-you |
| Developer access | Pay per call | DIY weights | Pay per call | Fernfly · free per call |
| Cost per query | High | Low | Low | Lowest |
A nonlinear model, on your hardware, inside a platform that retrains it.
Roadmap
From here
to Series A.
Today · 2026
Two products live
EverydaySeries + Fernfly in production. Fern-FC and Fern-Vision shipped.
6 months
Fern-Lang shipped
Native small general-purpose LLM. First enterprise device SDK in production.
12 months
Fern-Agent · Fern Pi OS
Native agentic reasoning shipped. Fern Pi OS beta image — Raspberry Pi wedge for the Fern OS line.
18 months · Series A
10+ paying customers
SMB and mid-market. ARR running rate to support the round. Repeat customers.
24 months+
Silicon
Custom ASIC for Fern models. Patents filed. FPGA validated.
Team
Built for this.
Founder · CEO · London
Dr Gaurav Gandhi. PhD nonlinear systems. Five years chip design at STMicro + Cadence. Prior founder experience. Royal Acad. of Eng. LIF Fellow.
CCO · London
Built, scaled, and exited tech businesses. Operational MBE. Defence and public sector procurement.
Head of Engineering · India
ex-Automattic, ex-Deel. 15+ years experience. Production systems at global scale.
Head of Research · London
Cambridge PhD, mathematics. Publication and IP.
Software Architect · India
MSc Applied Mathematics. 10 years building web and mobile products.
Full-stack Developer · India
BSc Computer Science. 10+ years full-stack experience.
Already in motion.
The ask
The ask.
One or two co-leads
Check email for exact ask. Or contact gaurav@nonlinear.technology
By Series A · 18 months
10+ paying customers across SMB and mid-market. Growing revenue. First device deployment live. ARR running rate to support a Series A.
Advanced Nonlinear Technologies Ltd · London
gaurav@nonlinear.technology