Skip to content
View CodeSageAnsh's full-sized avatar
🔨
Building computer-vision & LLM pipelines
🔨
Building computer-vision & LLM pipelines

Block or report CodeSageAnsh

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
CodeSageAnsh/README.md

Hi, I'm Anshuman Sarangi 👋

AI & automation engineer who ships into ERP. I build computer-vision and LLM systems for a manufacturing plant and wire them into SAP, so the output reaches the system the business actually runs on, not just a dashboard.

Developer (AI & Data Science) at Shaktigarh Textiles & Industries Ltd.

LinkedIn Email

🏭 What I've built in production

Project What it does Stack
shopfloor-ocr-sap Handwritten production sheet → LLM OCR → validation → SAP. In production, multi-mill. Geometric tilt correction, arithmetic cross-checks, computed confidence, human review. Python · Flask · Claude/Gemini · OpenCV · MongoDB · SQL Server · SAP GUI scripting · Telegram
weighbridge-truck-counter Counts trucks on a weighbridge from CCTV, with one timestamped screenshot per truck as audit evidence. Custom empty/full detector at about 97% mAP50. YOLO · OpenCV · Python
cctv-face-recognition-poc Feasibility study: 3,960-face index, 43 CCTV clips from S3. Showed by measurement why face attendance fails on wide-angle cameras and what would fix it. DeepFace · RetinaFace · Facenet512 · AWS S3
sap-automation-toolkit ABAP ALV routing report, plus a SAP GUI scripting audit that traced 2,103 material-text mismatches to a master-data defect. ABAP · Python · SAP S/4HANA MM

I also run a multi-camera RTSP → S3 recording pipeline on AWS EC2 (ffmpeg, shift-window scheduling, stall and zero-byte upload guards) and a daily camera vs HR attendance reconciliation.

🧰 Tools I use

  • Languages: Python · JavaScript (Node.js) · SQL · ABAP
  • AI / CV: LLM APIs (Claude, Gemini) · YOLO (Ultralytics) · OpenCV · DeepFace · scikit-learn · pandas
  • Backend & data: Flask · MongoDB · SQLite · SQL Server · REST APIs · Telegram / WhatsApp bots
  • Cloud & infra: AWS EC2 · S3 · Linux · ffmpeg · pm2
  • Enterprise: SAP S/4HANA (MM) · SAP GUI scripting · ABAP ALV

📌 How I work

  • I measure before I fix. A −2.6° camera tilt was the real cause of OCR row errors, not the prompt.
  • I don't trust a model's own confidence. Scores come from checks that can fail.
  • Nothing reaches the ERP without a human-in-the-loop review step and an audit trail.

🎓 B.Tech, Information Technology, KIIT University

Pinned Loading

  1. shopfloor-ocr-sap shopfloor-ocr-sap Public

    Handwritten production sheet -> LLM OCR -> validation -> SAP. Tilt correction, arithmetic cross-checks, computed confidence, human review.

    Python

  2. weighbridge-truck-counter weighbridge-truck-counter Public

    Counts trucks on a weighbridge from CCTV with a debounced line-occupancy counter and per-truck evidence screenshots.

    Python

  3. cctv-face-recognition-poc cctv-face-recognition-poc Public

    Feasibility study: face-recognition attendance on factory CCTV. 3,960-face index, 43 clips from S3, measured why it fails and what fixes it.

    Python

  4. sap-automation-toolkit sap-automation-toolkit Public

    ABAP ALV routing report and a SAP GUI scripting audit of material short text vs sales text.

    Python