JyothishGuru Enterprise Operations Hub

Monthly Recurring Revenue (MRR)

₹0

Loading live billing & tier telemetry...

Total Registered Users

0

Multi-tenant accounts

Saved Birth Profiles

0

Quota controlled

Active LoRA Checkpoint

base

Verified accuracy & checkpoint telemetry

Platform Health & Microsecond Caching

Vedic Ephemeris C-Bindings (`swisseph`) ● 100% Calibrated (Microsecond)

Deterministic Ephemeris Cache (Redis 7) ● Active (0.2ms Retrieval)

DeepMind-Class Google Research Harness² Recursion ● Active (k=1 Multi-turn)

Payment Processing Gateway (Razorpay HMAC-SHA256) ● Ready for UPI / Cards

Subscription Breakdown

Enterprise (Guru) 0% (0)

Pro (Jyotishi) 0% (0)

Free (Jigyasu) 0% (0)

Registered Users, Quotas & Balances

Identity & Email	System Role	Subscription Tier	Saved Profiles	Credits Balance	Operations

Dynamic Package Entitlements Switchboard

Enable or restrict astrological engines and calibrate profile quotas per tier without redeployment.

Active Canary Feature Flags

Provision Test User Cohort

Tester Discrepancy & Accuracy Telemetry

Tester / Cohort	Feature Key	Accuracy Rating	Discrepancy Reported	Comments & Shastric Observations

LoRA Model Checkpoints Registry

Hot-swap fine-tuned LoRA weights in GPU VRAM without restarting the server.

Automated & 1-Click Offline QLoRA Retraining Pipeline

● Ready (Weights Protected)

Curates verified seeker Thumbs-Up SFT turns (`user_feedback_sft.jsonl`), blocks any Thumbs-Down regressions, distills multi-probe Google Harness² wisdom (`k=3`), runs QLoRA (`Qwen2.5-VL-7B 4-bit NF4`), and enforces ChatML + Empathy quality gates before hot-swapping checkpoints.

Verified Thumbs-Up SFT

0 Turns

0 new since last retrain

Thumbs-Down Calibration

0 Flags

Excluded from SFT corpus

Harness² Wisdom & Distill

0 Lessons

0 distilled ChatML samples

Latest Evaluation Gate

Awaiting Run

Schema ≥98% • Empathy ≥80%

Pipeline Hyperparameters & Auto-Trigger Threshold Policy

QLoRA Epochs (`1`–`10`)

Harness² Distill Samples (`k=3`)

Feedback Upsample (`1x`–`5x`)

Auto-Trigger Threshold (new thumbs-up)

Default Auto-Trigger Mode

Auto-trigger background retrain when the new thumbs-up count hits the threshold Auto-promote & hot-swap LoRA checkpoint when Evaluation Gate passes

Stage: Initializing... 0%

Job ID & Time	Trigger / Mode	Merged Dataset	Evaluation Gate	Checkpoint	Status
No retraining runs recorded yet.

Vedic Training Lake & Verified Corpora

17 cataloged astronomical datasets including 15,807 historical horoscopes.

Dataset Filename	Category	Format	Size	Sample Count	Description

Real-Time Inference Hyperparameters

Inference Temperature: 0.35

Low (0.1-0.3) for mathematical determinism • High (0.6+) for philosophical prose

Top-P Nucleus Sampling: 0.90

Quantization Format

Google Research Harness² Recursion Rounds

Tailscale Split-Node & Remote GPU Worker

Hot-swap Tailscale GPU routing, worker secret, timeouts & fallback behavior live across VPS workers without restarting FastAPI.

Deployment Node Role (`JG_NODE_ROLE`)

Tailscale GPU Worker URL (`http://100.x.y.z:8090`)

Supports Tailscale CGNAT (`100.64.0.0/10`), `.ts.net` MagicDNS, or HTTPS tunnel.

Shared GPU Worker Secret (`X-GPU-Worker-Secret`) Clear stored secret

Tailscale RPC Stream Timeout (Seconds)

Recommended: 60–120s for long bilingual Qwen2.5-VL + TTS streams.

GPU Worker Max Concurrency (`GPU_MAX_CONCURRENCY`)

Semaphore guard preventing CUDA OOM under burst traffic.

Enable Remote Tailscale GPU Routing Auto-Fallback to Local CPU/Edge-TTS if GPU Unreachable

Cross-OS Byte Audio, Multimodal & Consultation Defaults

Preferred TTS Stream Format

Default Consultation Language

Default Kundli SVG Style

Acoustic Voice Emotion Detection (SenseVoice) Offline Chime WAV Mode (Air-Gapped / Zero WAN TTS)

Harness² Self-Improvement, Generation & Gateway Rate Limit

Max New Tokens (`128`–`4096`)

Temperature (`0.0`–`1.5`)

Top-P Nucleus (`0.05`–`1.0`)

Quantization Precision

Harness² Recursion Depth (`1`–`3`)

API Rate Limit (Requests / Min / IP)

Enable Google Research Harness² Task-Level Self-Improving Synthesis

Local-First Identity, Google OAuth (`localhost`) & Outbound SMTP

Google OAuth Client ID (`localhost`) Outbound SMTP Host (Optional) SMTP Port SMTP Username SMTP App Password From Email + Clear Stored Secret

YouTube Data API v3 & Instagram Reels (Graph API) Video Publishing

Open Video Generator Studio ↗

Configure OAuth2 / Graph API credentials to push multi-language Rashi + Lagna + Running Dasha Transit Magazine Videos directly to YouTube (16:9 HD & 9:16 Shorts) and Instagram Reels (9:16). Secrets are encrypted at rest and masked in the UI.

Enable Direct YouTube Data API v3 Video / Shorts Publishing Enable Direct Instagram Graph API Reels Publishing

YouTube OAuth2 Client ID YouTube OAuth2 Client Secret Clear stored Client Secret YouTube OAuth2 Refresh Token (`youtube.upload`) Clear stored Refresh Token YouTube Channel ID & Default Privacy

Instagram Business Account ID (`ig_user_id`) Instagram Graph API Long-Lived Access Token Clear stored IG Token Public HTTPS Media Base URL (Required by Meta Graph API for Reels) Default Video Aspect & Narration Language

Server Disk Folder: c:\Users\maitr\Documents\works\JyothishGuruAI\data\social_videos\

+ Add New Birth Profile (Unlimited Admin Storage)

Search all 722,000+ Indian villages, mandals, small towns (e.g. Tenali) & global places, or pin exact hospital/street coordinates on Google Maps.

Seeker Full Name *

Gender

Date of Birth *

Time of Birth (HH:MM:SS) *

Place / Town / Village of Birth *

Latitude (6-dec)

Longitude (6-dec)

UTC Offset

Ayanamsa

Saved Birth Profiles Registry (0)

Seeker Name & Gender	Date & Time of Birth	Birth Place & 6-Decimal Coordinates	Ayanamsa / Owner	Actions

Google Research Harness²: Recursive Self-Improving Engine

arXiv:2610.14224

Continuous recursive self-improvement without modifying underlying model weights (Zero Catastrophic Forgetting). Connects Task-Level Recursion (Propose–Probe–Compose via behavioral contrasts) with Harnessing-Level Recursion (Domain Experience Library across 8 life domains).

Self-Improving Active (k=1)

Execution Harness

8 Core Components

Prompt, Guardrails, Memory, Skills...

Task-Level Recursion

Propose–Probe–Compose

Contrasts k candidate executions

Experience Library R

0 Lessons

★ 0 Curated Gold Rules across 8 domains

Model Weight Status

Fixed / Untouched

Zero catastrophic forgetting

8 Modular Execution Harness Components

Domain Experience Library & Curated Gold Rules

Search cue-matched rules across all 15 traditions, inspect domain buckets, or author custom Acharya Gold Rules.

★ Gold Rules Only

Security Audit Trail & Credit Ledger

Event Timestamp	User Identity	Action Type	Credit Delta	Balance After	Reference / Audit Context

Executive KPI Dashboard

Platform Health & Microsecond Caching

Subscription Breakdown

Registered Users, Quotas & Balances

Dynamic Package Entitlements Switchboard

Active Canary Feature Flags

Provision Test User Cohort

Tester Discrepancy & Accuracy Telemetry

LoRA Model Checkpoints Registry

Automated & 1-Click Offline QLoRA Retraining Pipeline

Vedic Training Lake & Verified Corpora

Real-Time Inference Hyperparameters

Tailscale Split-Node & Remote GPU Worker

Cross-OS Byte Audio, Multimodal & Consultation Defaults

Harness² Self-Improvement, Generation & Gateway Rate Limit

Local-First Identity, Google OAuth (`localhost`) & Outbound SMTP

YouTube Data API v3 & Instagram Reels (Graph API) Video Publishing

+ Add New Birth Profile (Unlimited Admin Storage)

Saved Birth Profiles Registry (0)

Google Research Harness²: Recursive Self-Improving Engine

8 Modular Execution Harness Components

Domain Experience Library & Curated Gold Rules

★ Author / Edit Acharya Curated Gold Rule

Security Audit Trail & Credit Ledger

JyothishGuru Control Center

Platform Health & Microsecond Caching

Subscription Breakdown

Registered Users, Quotas & Balances

Dynamic Package Entitlements Switchboard

Active Canary Feature Flags

Provision Test User Cohort

Tester Discrepancy & Accuracy Telemetry

LoRA Model Checkpoints Registry

Automated & 1-Click Offline QLoRA Retraining Pipeline

Vedic Training Lake & Verified Corpora

Real-Time Inference Hyperparameters

Tailscale Split-Node & Remote GPU Worker

Cross-OS Byte Audio, Multimodal & Consultation Defaults

Harness² Self-Improvement, Generation & Gateway Rate Limit

Local-First Identity, Google OAuth (`localhost`) & Outbound SMTP

YouTube Data API v3 & Instagram Reels (Graph API) Video Publishing

+ Add New Birth Profile (Unlimited Admin Storage)

Saved Birth Profiles Registry (0)

Google Research Harness²: Recursive Self-Improving Engine

8 Modular Execution Harness Components

Domain Experience Library & Curated Gold Rules

★ Author / Edit Acharya Curated Gold Rule

Security Audit Trail & Credit Ledger

Admin High-Precision Google Map Picker