Experimental Research Build
Digitize Tamil with
Machine Precision.
"தேமதுரத் தமிழோசை உலகமெலாம் பரவும் வகை செய்தல் வேண்டும்"
Accelerate fine-tuning pipelines for classical Tamil literature and ancient Siddhar manuscripts using Rust parallelism and local Edge-AI.
cargo run --release -- --pdf-path ./siddhar.pdf
[19:22:27]Starting OCR process...
Batch processing...0%
AI Context-Aware Repair
Raw OCR (Tesseract)
சித்தர்க ள் நாதன் சிவயோக மாமு னி . ,
அகர மு தல எழுத் தெல்லாம் ஆ தி...
Gemma 4 Corrected
சித்தர்கள் நாதன் சிவயோக மாமுனி
அகர முதல எழுத்தெல்லாம் ஆதி...
Project Roadmap
The Future Timeline
Currently in an experimental research phase for Siddhar manuscripts. Here is where the architecture is heading.
Phase 1: Production Baseline
Completed
- • LLVM optimized Rust Engine.
- • Rayon Work-Stealing Pool.
- • Core Tesseract Binarization mapping.
Phase 2: Edge-AI Correction
Active Development
- • Local Inference via Ollama POST.
- • Gemma 4.5B Token Repair.
- Upcoming: Custom fine-tuned Gemma 4 QAT Model trained exclusively on Siddhar vocabulary.
Phase 3: Native Delivery
Planned
- • Tauri v2 Core Backend Wrap.
- • HTML5/TS responsive Frontend GUI.
- • IPC Commands for seamless bridging.
Deploy in Minutes
100% offline. Zero data sent to the cloud.
1
Install Dependencies
brew install pdfium tesseract tesseract-lang
2
Setup Local Edge AI
ollama run gemma
3
Clone & Compile Engine
git clone https://github.com/digital-tamil/tamil-simple-ocr.git cd tamil-simple-ocr cargo build --release
ALU Cycles & Registers
Hardware Logic Schematic
A real-time trace schematic of Rust Multi-threading passing state buffers into the AI co-processor.
INGRESS_BUS [0x7FF]
STATUS: PIPELINE_READY
D0சி
D1த்
D2த
D3ர்
TX LINE CLK: 12.4 GHZ
RAYON_ALU_COMPLEX
CORE_0
OP: TESS_BIN_LSTMCORE_1
OP: TESS_BIN_LSTMCORE_2
OP: TESS_BIN_LSTMCORE_3
OP: TESS_BIN_LSTMPKG_0 SYNC_LOCK
GEMMA_COPROC_NPU
CONTEXT_WINDOWVRAM: 8.4 GB
INFERENCE RATE148 T/S
BOARD_REV: 0xDEADBEEFCOMP_LEVEL: OPT_3_LTO_TRUECLOCK: SYSTEM_REF_PCLK_12