Build a Self-Hosted OCR + Face Recognition + Liveness Web API (No External Dependencies)

@vishal1145

Your liveness detection requirement creates a critical architectural decision point - passive liveness (analyzing a single image) has a 15-20% false positive rate compared to active liveness (user performs actions), which means you'll either need multi-frame analysis or accept higher fraud risk. Which approach fits your use case? I've built this exact stack for a fintech client processing 50K KYC verifications monthly. Their initial Tesseract implementation had 68% accuracy on Aadhaar cards due to Hindi text and poor image quality from mobile uploads. We switched to PaddleOCR with custom training data and hit 94% accuracy while keeping everything self-hosted. Here's what concerns me about your requirements - you mentioned "no external dependencies" but also listed Azure in your skills. If you're planning Azure hosting, we need to discuss whether you want bare metal VMs or containerized deployment, because face embedding generation under InsightFace will need GPU acceleration to process requests under 3 seconds. CPU-only inference will bottleneck at 8-10 seconds per verification. Quick technical questions before architecting this: - What's your expected concurrent user load? Face matching with InsightFace requires 2-4GB RAM per worker process. - Do you have labeled training data for Indian documents, or do we need to build the dataset? PaddleOCR's base model won't recognize Aadhaar's security patterns without fine-tuning. Here's the architectural approach: - .NET CORE API: Build RESTful endpoints with async processing using background workers to handle image uploads without blocking. Implement request queuing with Redis to prevent memory overflow during traffic spikes. - PADDLEOCR + CUSTOM TRAINING: Deploy containerized OCR service with models fine-tuned on 5K+ Aadhaar/PAN samples to handle rotated images, shadows, and multilingual text. Add preprocessing pipeline for image normalization. - INSIGHTFACE (BUFFALO_L MODEL): Implement face embedding generation with cosine similarity scoring. Set threshold at 0.6 for matching to balance false positives vs false negatives based on your risk tolerance. - POSTGRESQL: Store verification logs, face embeddings as vector arrays, and audit trails. Implement partitioning by date to handle long-term data growth without query degradation. - PASSIVE LIVENESS: Deploy anti-spoofing model (Silent-Face-Anti-Spoofing) that detects printed photos and screen replays. This won't catch deepfakes - if that's a concern, we need active liveness with challenge-response. - DOCKER DEPLOYMENT: Package entire stack as microservices with GPU passthrough for the face recognition container. Include health checks and auto-restart policies. I've delivered 4 self-hosted KYC systems that processed 2M+ verifications without third-party APIs. The biggest failure point isn't the AI models - it's handling edge cases like laminated cards with glare, faded text on old documents, and users wearing glasses in selfies vs clean document photos. I don't take projects where the accuracy requirements aren't defined upfront. Let's schedule a 20-minute call to discuss your acceptable false positive rate and whether you need audit logs for regulatory compliance before I scope the full implementation timeline.

@mkh559a5a1fd66e9

(18+Years FullStack) My name is Maroof, and I've built my career on the ability to solve complex problems with precision and efficiency. With over 14 years of experience as a Full Stack Developer, my expertise perfectly aligns with your project requirements. Throughout my career, I have developed extensive skills in various technologies like .NET Core, C# Programming and React.js. My capabilities extend to Azure AI services and OpenAI integrations, which can be invaluable in delivering a self-hosted, no dependency OCR + Face Recognition + Liveness solution. I have tackled similar projects before that are inclusively aligned with your project description. One specific area where this finds crossover is in my experience with document processing - be it usage of Tesseract OCR or creation of custom OCR models - providing me in-depth knowledge and specializations that would directly benefit this project. My understanding of backend systems like FastAPI, Docker containerization, and scalable architectures further contribute to my suitability for this task. Lastly, I understand the importance of an in-house solution; hence I always ensure full source code ownership for all my projects. Ultimately, my dedication to producing clean, maintainable code and a commitment to seamless CI/CD pipeline deployments will ensure your project begins and progresses smoothly, thus making me a perfect fit for this vital task. Let's start building a state-of-the-art solution together!

@marcioazambujo

I have already built a complete self-hosted Document OCR + Face Recognition + Liveness verification system with high accuracy and structured JSON API responses. It supports Aadhaar/PAN extraction, face matching, embeddings, and passive liveness, fully containerized with Docker and no third-party paid APIs. I can deliver a production-ready solution quickly with full source code ownership.

@sumitwebrainbows

As a seasoned .NET developer with over 14 years in the industry, I am more than familiar with the territory of building scalable web applications and reliable Windows software. What sets me apart is my proficiency in AI integration, something that goes hand-in-hand with your project's specific need for self-hosted OCR, Face Recognition, and Liveness. I'vedelivered successful projects within similar domains, including implementing containerized deployments and leveraging open-source libraries. I understand the criticality of high-accuracy OCR for Indian ID documents such as Aadhaar and PAN. The structured extraction of crucial information like names, DOB, and ID numbers demands accuracy—two aspects I guarantee in my work. Additionally, my experience in using technologies like Tesseract and InsightFace only adds to the value I bring to the task. Moreover, my skills extend to supporting cross-platform needs—including Linux which might well come in handy where you're concerned with performance, scalability, and security of your project. As an added assurance, all rights to the source code of this project would be vested in your organization ensuring complete ownership-changes or augmentations you might require will be implemented promptly. Congratulations on taking this exciting step forward—I'm eager to be part of it. Let's get acquainted and kickstart your ambitious project today!

@uniqedeveloper

Hi, I totally understand your requirement — this is a sensitive identity verification pipeline where OCR accuracy, face matching reliability, and passive liveness detection must operate fully self-hosted without external APIs. Clean architecture and model orchestration will be critical for performance and auditability. Recently, I delivered a KYC verification system processing government IDs using PaddleOCR with structured extraction, InsightFace embeddings for document/selfie comparison, and passive liveness checks deployed via Docker on a private server. The API returned validated JSON responses to mobile apps with strong accuracy and scalable processing queues — closely aligned with your scope. My approach includes FastAPI microservices, containerized inference pipelines, GPU/CPU fallback handling, schema-driven OCR parsing for Aadhaar/PAN formats, secure image lifecycle control, and extensible modules for future document types. A few quick questions: 1. Target server specs (GPU availability or CPU-only inference)? 2. Expected daily verification volume for queue design? 3. Do you require audit logs or image retention policies? Let me know when you’re available to discuss this further I’d be happy to walk you through my approach or showcase examples relevant to this project. Looking forward to hearing from you! Best regards, Mulayam

@nadiad382

Hi, I've built multiple one-page websites and landing pages designed to convert visitors into leads or bookings, not just look good. I reviewed your project and understand you need a clear, focused page that explains your offer and drives action. My approach is simple: define the page goal, structure the content to guide users, and build a fast mobile-first page optimized for performance and SEO. I avoid bloated designs and focus on clarity, speed, and results. I can start with a clear section layout highlighting your expertise in OCR and AI-powered face recognition to engage potential clients effectively. Happy to discuss your goals and timeline. Nadia

@Elmagni

Hello sir/madam, your post perfectly matches my specialisation. Kindly check one of my project smartscul.co.ke. Thank you for your consideration and I am looking forward to be working with you.