Siddha

The Siddha Lab

Grow a model.
Watch it think.

This is not a video of our product — it is the method, running in your tab. Because Reverse Synthetic Networks are closed-form, a faithful miniature of the embedding-only mode synthesizes below in milliseconds, entirely client-side. Nothing you type is sent anywhere.

Synthesis Studio · zero gradients, zero servers
Step 1 · The soil

Labeled examples

billing · 12 technical · 12 sales · 12
billingI was charged twice for my subscription this month, please refund the duplicate charge
billingMy invoice shows the wrong amount and the tax number is missing from the invoice
billingHow do I update the credit card used for my subscription payment
billingThe subscription renewed automatically but I wanted to cancel before the payment date
billingPayment failed but the amount was still deducted from my card
billingI need a receipt and invoice for last month's subscription payment
billingWhy did the price of my subscription increase on the latest invoice
billingPlease cancel my subscription and refund the unused amount to my card
billingThe refund you promised has not appeared on my card statement yet
billingI was charged in dollars but my invoice should be in rupees, please correct the charge
billingYour system charged my card twice and the duplicate payment needs a refund
billingCan I switch from annual to monthly payment without losing the subscription
technicalThe app crashes every time I try to upload a file from my phone
technicalLogin keeps failing with an error message after the latest app update
technicalThe dashboard will not load and the page shows a network error
technicalExport produces a corrupted file that will not open in Excel
technicalNotifications stopped working on my phone after the update
technicalThe API returns an error when I send a large request, looks like a server bug
technicalVerification codes never arrive so I cannot login to the app
technicalThe charts render with a display bug in dark mode, labels are invisible
technicalSync between the app and the web dashboard is broken, data does not load
technicalGetting a server error every time I save, the bug started after the update
technicalPassword reset email never arrives and login fails with an unknown error
technicalFile upload hangs at ninety percent and then the app crashes
salesDo you offer a discount on the enterprise plan for educational teams
salesWe are a team of forty people, which plan and pricing would you recommend
salesCan I get a demo of the enterprise features before purchasing a plan
salesWhat is the pricing difference between the pro plan and the business plan
salesIs there a partner program with special pricing for agencies
salesWe need an on-premise deployment, who can I talk to about enterprise pricing
salesDoes the enterprise plan include a dedicated account manager and support
salesPlease send a quote for two hundred seats on the business plan
salesInterested in upgrading our team plan, can someone from sales contact me
salesWhat discount do you offer for a multi-year enterprise contract
salesCan we schedule a demo and discuss pricing for our team next week
salesLooking for a quote comparing the pro and enterprise plans for fifty seats
Step 2 · Growth

Synthesize the model

One closed-form pass: token–class statistics, IDF, discriminative weighting, class prototypes. No gradients — which is why it can run here, in your tab. Your data never leaves this page.

Step 3 · The harvest

Classify new text

A faithful miniature of RSN's embedding-only mode — the label-enriched token statistics our research identified as the primary small-scale signal — not the full four-pillar system. The full architecture adds spectral interconnection, synthesized attention, and ensemble convergence on real workloads.

New · live vision demo

Grow a digit-recognition network. Then draw for it.

1,797 real research samples, synthesized into visible prototype neurons in your tab — then it reads your handwriting and shows you exactly which neuron it matched, next to an honest comparison with a trained neural network.

Open the digit demo

Benchmark explorer

The numbers — wins, losses, nulls.

Three tabs. One is flattering, two are not. All three are the same size, because that is the deal we made with the evidence.

20 Newsgroups (4-class)1,732 train / 1,126 test docs · synthesis ≈60 s CPU
Siddha RSN — zero training (ensemble, 9/10 seeds above baseline; 86.4% single-model)87.0% ± 0.5%
TF-IDF + tuned linear SVM86.1%
Topical text (5-class)105 train / 45 test docs · synthesis ≈15 s CPU
Siddha RSN — zero training (embedding-only synthesis)68.9%
TF-IDF + cosine60.0%
Digit recognition (8×8)1,437 train / 360 test images · synthesis < 1 s CPU
Siddha RSN — zero training (zero gradient steps)90.8%
MLP trained with backpropagation96.4%

Accuracy, higher is better. The digits row shows the trained MLP ahead — synthesis loses there and we chart it at full size.

Energy estimator

What your cadence costs the grid.

Set how often you'd rebuild a model. The assumptions are printed beside the result, and the scope of the claim beneath it.

30rebuilds / month

High-cadence teams rebuild daily or hourly as data drifts. Drag the slider to your cadence; the arithmetic updates from the stated assumptions below.

0.02 kWh
Siddha synthesis energy
60 s · ~35 W CPU per rebuild
24 kWh
Equivalent GPU fine-tunes
2 h · ~400 W A100 per rebuild
1,371×
Energy ratio
GPU path ÷ synthesis path
17 kg
CO₂e avoided / month
≈ 142 km not driven

Stated assumptions, honestly scoped: synthesis at 60 s / ~35 W (measured order of magnitude for the 20 Newsgroups ensemble); fine-tune at 2 h on one ~400 W A100; grid intensity 0.71 kg CO₂e/kWh (Indian grid average, CEA). The comparison applies to the small discriminative workloads where synthesis is competitive — it is not a claim about LLM training, which synthesis cannot replace.