AI/ML Engineer & Developer
ANISH BHAT
Building intelligent systems with Generative AI, Agents, Automation, and Product Development.
About Me
AI/ML Engineer and backend developer with hands-on experience designing scalable backend systems, API-driven architectures, and production-grade ML pipelines.
Proficient in Python, OOP, and clean code practices with a strong emphasis on writing test cases, debugging production issues, and maintaining reliable version control workflows.
Proven track record of shipping end-to-end systems—from architecture design through testing and deployment—with measurable impact on performance, reliability, and business outcomes.
Academic Background
Bachelor of Engineering in Computer Science (Artificial Intelligence)
KLE Technological University, Hubballi, Karnataka
Pre-University (PCMB)
Expert PU College, Mangaluru, Karnataka
Honors & Certs
Certifications
Awards
1st Place, College Mechatronics Competition (130+ teams)
Hackathon Finalist & Team Lead, Smart India Hackathon and HackKarnataka
Technical Arsenal
AI & Machine Learning
Frameworks & Libraries
Backend & Web
Databases & Cloud
Professional Experience
Freelance AI Engineer (Lead Developer)
- ▹Led development of a real-time vision-based assistive system using object detection and scene understanding for visually impaired users.
- ▹Designed optimized computer vision inference pipelines using OpenCV and deep learning models, achieving 2.5× faster inference on Raspberry Pi through INT8 quantization.
- ▹Integrated camera-based perception with Firebase APIs and Android applications for real-world deployment.
- ▹Containerized complete vision pipelines using Docker and implemented CI/CD workflows, reducing deployment time by 35%.
- ▹Led and coordinated a 3-member engineering team, ensuring on-time delivery of embedded AI milestones.
Embedded AI Intern
- ▹Trained and evaluated YOLOv11 object detection models on KITTI dataset using statistical metrics and validation techniques, achieving over 65% model size reduction through structured pruning and INT8 quantization.
- ▹Performed data preprocessing and feature engineering on autonomous driving datasets to improve detection accuracy across Raspberry Pi and Jetson Nano edge devices.
- ▹Built a scalable backend ML pipeline integrating real-time object detection, lane detection, and sensor fusion with end-to-end latency of approximately 210ms; debugged latency bottlenecks and sensor sync failures to meet production performance targets.
- ▹Deployed optimized deep learning models on embedded hardware, implementing REST APIs for telemetry and system monitoring; managed codebase on GitHub with clean branching and documented commit history.
Svarra
Multi-Agent AI Voice Automation Platform handling autonomous inbound/outbound voice calls at scale.
Problem Solved
Automating complex conversational workflows like lead qualification, appointment booking, and follow-ups securely.
Key Outcomes
- ▹ Exposed modular REST APIs for 4 specialized agents.
- ▹ Implemented idempotent webhook handling for CRM updates and calendar bookings.
- ▹ Delivered a real-time analytics dashboard tracking call duration, cost, and conversion rates.
- ▹ Enabled multi-lingual voice interactions for the Indian market.
Offline P2P Blockchain Wallet
A decentralized offline peer-to-peer blockchain payment system that enables secure financial transactions without internet connectivity, bridging the gap for rural and underserved communities.
Problem Solved
Financial inclusion in India remains a challenge where internet connectivity is unreliable. Traditional digital payments exclude millions. This solution prevents double-spending offline and ensures trustless transactions without continuous internet access.
Key Outcomes
- ▹ Built a secure, truly offline-first decentralized payment system
- ▹ Eliminated reliance on continuous internet connectivity for rural users
- ▹ Solved offline double-spending via local nonce tracking and on-chain validation
Embedded ADAS Prototype
Designed an end-to-end embedded Advanced Driver Assistance System integrating realtime object detection, lane detection, ultrasonic sensing, and motor control.
Problem Solved
Building an affordable and deployable ADAS prototype on low-power edge hardware.
Key Outcomes
- ▹ Achieved ~6 FPS real-time object detection inference on a Raspberry Pi.
- ▹ Integrated ultrasonic sensors and motor control for real-world interaction.
- ▹ Developed a Flask-based web interface for real-time visualization, telemetry, and experimental evaluation.
YOLO-OptiMob
Built a complete optimization pipeline for deploying YOLO11 object detection models on resource-constrained edge devices.
Problem Solved
Massive memory footprints and slow inference times of default object detection models on mobile devices.
Key Outcomes
- ▹ Selected 40% L1 unstructured pruning as optimal trade-off.
- ▹ Reduced model size from 11.4 MB to 2.5 MB.
- ▹ Successfully converted to TFLite for deployment on Android-based edge platforms.
Vision-Language Scene Understanding (BLIP)
Developed a real-time vision-language system using BLIP to generate natural language scene descriptions from live visual input.
Problem Solved
Enabling semantic, natural language reasoning over complex visual scenes rather than just bounding boxes.
Key Outcomes
- ▹ Generated structured textual representations of scenes to enable downstream reasoning by large language models.
- ▹ Achieved 82% accuracy through dataset curation, fine-tuning, and inference optimization.
Agentic RAG Medical Data Extractor
Designed a medical document RAG pipeline using Gemini embeddings, Pinecone, Firebase, and GCP APIs.
Problem Solved
Extracting highly specific, structured summaries from massive, unstructured medical PDFs securely.
Key Outcomes
- ▹ Extracted structured summaries from unstructured medical PDFs for research use.
- ▹ Automated the workflow reliably via Docker and n8n.
Elderly Care Full-Stack Platform
Developed a full-stack platform enabling ambulance booking, medicine ordering, and elderly care services.
Problem Solved
Providing an accessible, all-in-one care hub for elderly patients to manage their critical services.
Key Outcomes
- ▹ Delivered a robust, real-time application for medicine and ambulance orchestration.
- ▹ Secured platform with Firebase Auth.
Research Publications
View All PapersLongDocAI: A Quantized Modular Pipeline for Multimodal PDF Summarization and QA
The rapid expansion of scientific literature has made it increasingly difficult for researchers to extract meaningful insights from dense academic documents. We introduce LongDocAI, a multimodal, multi-model framework designed to efficiently understand and interact with scholarly PDFs at scale. The system combines Donut, an OCR-free document parser, with a hybrid summarization module based on BARTlarge- CNN, trained respectively on over 200,000 paper-summary pairs and 30,000+ question-answer examples. To ensure faster and resourceefficient inference, we apply LLM.int8 post-training quantization across all transformer models, significantly reducing memory usage and latency without compromising output quality. LongDocAI is built as a unified pipeline that handles both layout-aware parsing and semantic-level understanding through summarization and interactive question answering. In our evaluations, the system achieved a 15% improvement in ROUGEL, a 28% boost in METEOR, and a QA accuracy of 85.3%, based on expert human assessments. Quantization further led to a 70% reduction in model size and a 40% decrease in inference time, making the system suitable for real-time or edge deployment. By integrating multimodal document understanding, summarization, question answering, and quantization into a single, modular pipeline, LongDocAI offers a scalable, lightweight solution for navigating and understanding long-form academic texts—benefiting researchers, reviewers, and knowledge platforms alike.
Embedded Intelligent ADAS Car Prototype Using Raspberry Pi and YOLOv12n
This paper presents the design, quantization, and deployment of an embedded Advanced Driver Assistance System (ADAS) prototype on a Raspberry Pi 3, utilizing a custom-trained YOLOv12n object detection model. The system achieves real-time multi-class object detection and autonomous control of a test vehicle within the resource constraints of low-power edge devices. To optimize performance, posttraining INT8 quantization was applied to YOLOv12n, resulting in over 65% reduction in model size and nearly 3× faster inference, with minimal accuracy loss. Object detection outputs are combined with ultrasonic sensor measurements to enable obstacle-aware vehicle control, including braking and navigation. Lightweight classical computer vision methods facilitate lane detection for lane-keeping functionality. Additionally, a Flask-based dashboard streams detection overlays and telemetry data for monitoring. The deployed system operates at approximately 6 frames per second on the Raspberry Pi 3 with a responsive control latency of around 210 ms, demonstrating the practical feasibility of deploying deep learning–based ADAS functions on low-cost, resource-constrained embedded platforms. This work highlights the effectiveness of quantization techniques in enabling real-time perception for embedded autonomous driving applications.
YOLO-OptiMob: A Pipeline for Optimizing YOLO11 Models for Edge Deployment
This paper introduces YOLO-OptiMob , a comprehensive pipeline for optimizing YOLO11 models for deployment on edge devices. The process begins with creating and preprocessing a custom dataset featuring seven object classes: bike, car, cat, dog, person, handbag, and water bottle. The YOLO11 model is trained on this dataset and optimized using L1 unstructured pruning, with rates of 30%, 40%, and 50% evaluated. Based on the results, a 40% pruning rate was selected as it offered the best balance between model size reduction and accuracy retention. Post-training INT8 quantization further compresses the model, reducing its size from 11.4 MB to 2.5 MB. The optimized model is then converted into TensorFlow Lite (TFLite) format, ensuring compatibility with Android-based edge devices. This work presents a practical pipeline for efficiently adapting YOLO11 to resource-constrained environments, achieving significant size reduction while maintaining high detection accuracy.
Let's Build
Whether it's a freelance AI pipeline, a scalable backend, or a full-stack product, I'm ready to collaborate.