Our First Launch - Gahna 0.1B Scheme Eligibility Checker

Domain: Government Welfare and Scheme Access
Tagline: Sarkari Yojanaayein Mutthi Mein (Government Schemes in Your Hands)

GAHNA 0.1B will mark the beginning of a new era in sovereign AI - one where domain-specific intelligence is delivered at scale with minimal compute, maximal control, and verifiable precision. This compact, production-grade Small Language Model (SLM) is engineered to instantly determine which central and state government schemes a citizen is eligible for, based on structured demographic, socioeconomic, and regional inputs - without relying on massive LLM infrastructure or opaque APIs.

Unlike generative LLMs, Gahna 0.1B is architected for deterministic, rule-aligned classification, trained on synthetically generated profiles grounded in real government eligibility frameworks. It offers zero hallucination, consistent performance, and field-specific interpretability - essential for use in governance applications.

🔹 🔹 Technical Summary:
🔹 Model Size ~100 Million Parameters - optimal for structured classification tasks and real-time eligibility reasoning
🔹 Architecture - Custom transformer optimized for field-token attention, sparse activation, and lightweight deployment
🔹 Input Format - Structured citizen profile: age, gender, income, caste, education, state, etc.
🔹 Training Corpus ~100M tokens, synthetically generated and rule-labeled based on policy trees, with edge-case augmentation
🔹 Inference API RESTful FastAPI microservice with <100ms latency on CPU
🔹 Deployment Footprint Runs on <8GB RAM; supports CPU-only and quantized edge deployment
🔹 Serving Format Dockerized, ONNX-exportable, and inference-aware for both cloud and offline mobile environments

Key Differentiators

🔹 No hallucinations, no free-form prompts - fully controlled, explainable outputs via structured API calls
🔹 Central + State scheme coverage, with modular extension for region-specific customization
🔹 Mobile-first design - deployable on low-resource Android devices, kiosks, or edge gateways
🔹 Secure by default - no external API calls, no data leaves infrastructure boundaries
🔹 Language agnostic architecture, enabling easy multilingual tuning (Hindi, Marathi, Bengali, Tamil, etc.)

Coming Next in the Gahna-GOV Family

Following the successful rollout of Gahna-GOV-0.1B, our roadmap expands into targeted adaptations to ensure inclusivity, localization, and last-mile accessibility across India’s diverse governance landscape.

🔹 🔹 Gahna-GOV-HIN – Hindi-First Scheme Eligibility Model
A multilingual version of the Gahna model trained specifically for Hindi-language inputs and outputs. Citizens will be able to query scheme eligibility using natural Hindi phrases or structured fields in Hindi.
🔹Incorporates Devanagari tokenization, colloquial variants, and region-specific vocabulary
🔹 Trained on transliterated and bilingual profiles from Hindi-speaking states
🔹 Built for integration with state portals, chatbots, and IVR systems

🔹🔹 Gahna-GOV-MAR – Maharashtra-Centric Custom Model
A state-optimized SLM tailored for Maharashtra’s central and regional welfare schemes, subsidies, and benefit eligibility frameworks.
🔹 Covers state-level programs like Shiv Bhojan, Panlot Vikas Yojana, Savitribai Phule Scholarship, etc.
🔹 Uses Marathi-aligned token sets and localized demographic markers (e.g., taluka, caste groups, occupation codes)
🔹 Supports language toggling, district-wise rule variations, and dynamic scheme updates via version control

🔹🔹Gahna-GOV-OFFLINE – Edge-Deployable Mobile Model
A fully quantized, low-footprint variant of the Gahna model designed for offline usage in Gram Panchayats, CSCs, and field service devices.
🔹 Runs efficiently on Android phones, tablets, and low-power edge hardware
🔹 Requires <200MB storage, <1GB RAM, and performs real-time inference without internet
🔹 Perfect for areas with low or no connectivity, enabling scheme access at the last mile