Nouman Chohan
Ex-Founder | AI researcher/Engineer | AI Pipeline Architect
Bangalore
Summary
A physicist at heart, a builder by nature, and an AI founder by choice — I scale ideas into real-world impact. As Co-Founder & Head of AI at Neurify, I took a vision rooted in core AI innovation and transformed it into a revenue-generating, investor-backed product with a production-grade pipeline (AnyVoice). From LLMs fine-tuned for low-end GPUs to multi-modal CV systems that function in the wild, I lead with a deep understanding of AI’s mathematical and architectural backbone. I’ve raised funds, built teams, won pitchathons, trained future engineers, and launched intelligent systems that don’t just run — they scale, adapt, and perform. Every line of code, every decision, is driven by a passion to turn abstract intelligence into applied, elegant engineering.
Overview
3
3
years of professional experience
3
3
Languages
Work History
Co-Founder & CEO / Head of AI
Neurify Technologies Pvt Ltd
09.2024 - 06.2025
- Started a company with a grand vision of innovating on the foundational core of AI.
- Built a team capable of handling each responsibility.
- Worked on giving the company a set direction and roadmap to follow.
- Worked on setting up the marketing and sales infrastructure to generate revenue within the company.
- I also worked as the Head of AI to build and train a team capable of building pipelines at the complete core of AI.
- Led the R&D and research team in optimizing the inference of LLMs on low-end GPUs and increasing the overall concurrency and scalability for the users.
- I worked closely on the business aspect of the company and managed to raise sufficient seed funds for the company.
- Researched enough to find the suitable market gap Neurify can solve. Built the first product unique enough to fill gap and also push company in realising it's vision gradually.
- Utilized the funds in a structured way to help scale the product, pump enough resources for R&D, create two separate sales channels (for products and services), and extensively put my time into training AI engineers in building scalable pipelines.
- Successfully managed to build a production-grade, in-house pipeline for the product AnyVoice.
- Won several Pitchathons as a standalone speaker to attract investors and potential clients. Good for branding.
- Worked closely at frequent intervals to identify gaps wherever necessary to find pivot potential, if required.
- I carried out extensive personal research on potential verticals to find the perfect market fit for our product.
- Personally led a talent acquisition department to only hire candidates with potential to grow exponentially based on custom rubrics decided by me.
AI Engineer/Researcher
Freelancer
01.2023 - 09.2024
- Led end-to-end fine-tuning of large open-source Transformer models (LLaMA 2, Falcon -40B, Bloom, Mistral, Dolly 2.0) using Hugging Face Transformers, PEFT (LoRA/AdapterFusion), and DeepSpeed ZeRO-3 for 3 A100 multi-GPU clusters.
- Applied quantization (GPTQ, bitsandbytes 8-bit/4-bit) and mixed-precision (FP16/FP8) training to reduce VRAM footprint while preserving model quality; integrated pruning (SparseFineTuning) to remove less than 10% of attention heads without accuracy loss.
- Architected scalable RAG systems using Haystack and LangChain, combining open-source LLMs with vector databases (FAISS, Milvus, Pinecone). Ingested multi-modal datasets (PDFs, Markdown, HTML, image OCR) via LangChain document loaders and Tesseract/EasyOCR pipelines.
- Built end-to-end document embedding pipelines: extracted embeddings with SentenceTransformers (all-MiniLM-L6-v2, CLIP-Text), indexed via FAISS GPU, and orchestrated similarity search (IVF PQ, HNSW) with real-time retrieval latencies of less than 30 ms.
- Developed multi-modal pipelines combining CLIP (ViT-B/32) and fine-tuned LLMs to perform Visual Question Answering (VQA), Image Captioning, and Visual Grounding. Utilized PyTorch Lightning to train a hybrid CLIP+LLaMA architecture (token-level cross-attention) on custom datasets (COCO, Flickr30k, and domain images).
- Designed pipelines for continual learning and parameter-efficient fine-tuning (LoRA, QLoRA) to update LLMs on streaming data (financial reports, technical documentation) without catastrophic forgetting.
Tech Lead
CloudOne AI Robotics Pvt Ltd
06.2023 - 08.2024
- Built an OCR engine using several deep learning, open-source models from Hugging Face. e.g.: Microsoft Table Transformer, YOLOv8, custom algorithms built using OpenCV and Dlib.
- Used an open-source multilingual text extraction model from Hugging Face, fine-tuned it to better extract Japanese handwritten text and numbers. Used Llama 3.1 for the context understanding of the text. Optimized the inference speed and concurrency for Llama 3.1 by hosting it on a cluster of Nvidia LS40s.
- Structured the above into several microservices that run asynchronously, which enhances user experience with reduced response time, even on a CPU. Used Celery, Redis, Docker, and Django.
- Built a stock-price prediction algorithm and deployed it for production-grade use that can trade stocks with a sure profit of 2% to 4%.
- Used FinRL and custom trading strategies to predict the correct entry point, and the corresponding exit point with a success rate of 6/10.
- Deployed it on AWS for production grade using Celery, Docker, Redis, and Django (for API communications).
- Built deep learning/computer vision desktop software that can analyze sports data based on video input. The software does not require traditional broadcast cameras to generate useful analysis. Any video input taken using dynamic PTZ cameras is sufficient to generate useful analysis.
- Used cutting-edge detection algorithms like Yolov7 (fine-tuned with a custom object collection dataset) for detection. Used DeepSORT + CSRT for tracking objects accurately. Used techniques like automated landmark detection and homography transformation to track the field and convert player coordinates to infer real-time insights from the data. For example, speeds, distance covered, pass probabilities, etc. (pitch perfect down to two decimal points).
Machine Learning Engineer
Critical AI Pvt Ltd
06.2022 - 07.2024
- Fine-tuned YOLOv7 for cutting-edge multi-category object detection.
- Integrated virtual fencing and optimized the model to work with low latency on the CPU using multi-processing and OpenVino.
- Integrated the service with a production-grade pipeline that can extract feed from multiple cameras and apply the segmentation and detection asynchronously using RabbitMQ, Docker, and Celery.
- Worked on building the algorithms for filtering noise from the blurred images, with low-resolution cameras. Applied the filtering with detection algorithm and integrated successfully with the pipeline.
- Built machine learning APIs to integrate with web-based applications and desktop-based applications.
- Built machine learning applications that were to be deployed in remote areas, which would work on the feed from PTZ IP cameras.
- Worked on applying machine learning to QGIS and DEM (digital elevation models) for offline location mapping without the help of GPS.
- Worked on collecting and cleaning sensor data from the gyroscope and accelerometer, and applied machine learning algorithms to the structured data to work with QGIS (DEM) simultaneously.
- Worked on making a standalone desktop application used for security purposes on-site using React.js, Node.js, and Electron, which communicates with the biometric sensor, collects the data, structures it, stores it in a separate database, and generates reports as per user requirements.
Education
Bachelor of Technology - Information Technology
MIT
Pune 04.2001 -
Skills
Pytorch / Tensorflow / Keras
FastAPI/Django (Mostly used FastAPI)
Deep Computer vision - MMdetection, OpenCV, Dlib, Yolov7-10, Vision transformers (ViT - swit, SAM), Pose Estimation, Optical Flow Estimation, Image super resolution (ESRGAN, ehancing image qualities using GAN's), Homographic transformation, Deep Tracking (DeepSORT, SORT) etc
LangChain, SGLang, RAG, CAG
Vector DB - Chrome, FAISS, pgvector
Brokers - Redis, RabbitMQ
FineTuning - Unsloth, Hugging Face Transformers PEFT, Colossal-AI, vLLM, OpenLLM (For serving), LLaMa Factory
Networking - NGINX/Apache web server configuration, Reverse proxying HTTP and WebSocket traffic, Configuring SSL termination and HTTPS redirection, Serving FastAPI, Flask, or LLM inference APIs behind NGINX
Nvidia NiM, Riva
Accomplishments
- Managed to establish a product-based AI company, build the product, scale it, raise funds, utilize it better to generate sales, and bring in revenue inside the company
- Created a funnel to hire exceptionally potential AI engineers, and set my own evaluation metric to spot them quicker and faster
- Trained a team of more than three AI engineers personally to build AI pipelines with microservice architecture from scratch, and make them production-ready and scalable
- Completed a project for the Indian Army under a private organization with near human accuracy.
- Deployed several fine-tuned, scalable SLMs and LLMs to sustain a heavy amount of traffic with good feedback
Physics Nerd
I've always been deeply fascinated by how everything works — from the motion of galaxies to the behavior of particles at the quantum level. Physics, both classical and quantum, has been my lens to explore these fundamental truths. I see physics not just as a subject, but as a language — one that helps us decode the underlying principles of the universe. This lifelong curiosity has shaped how I think, approach problems, and design solutions. Whether I'm building AI systems or optimizing complex architectures, I bring the same analytical rigor and desire to understand the 'why' behind the 'how'. It's this mindset — driven by a nerd-like obsession with cause and effect — that fuels my ability to learn deeply, adapt quickly, and create systems that are not just functional, but elegant and efficient.
Timeline
Co-Founder & CEO / Head of AI
Neurify Technologies Pvt Ltd
09.2024 - 06.2025
Tech Lead
CloudOne AI Robotics Pvt Ltd
06.2023 - 08.2024
AI Engineer/Researcher
Freelancer
01.2023 - 09.2024
Machine Learning Engineer
Critical AI Pvt Ltd
06.2022 - 07.2024
Bachelor of Technology - Information Technology
MIT
04.2001 -
Similar Profiles
Venkata Nagaraju InakolluVenkata Nagaraju Inakollu
Technical Lead(System Design) at Relysys Technologies Pvt Ltd (now Invendis Technologies pvt ltd)Technical Lead(System Design) at Relysys Technologies Pvt Ltd (now Invendis Technologies pvt ltd)
S ChetanS Chetan
Assistant Manager - Finance at ANIMALL TECHNOLOGIES PVT LTD, Technologies Pvt LtdAssistant Manager - Finance at ANIMALL TECHNOLOGIES PVT LTD, Technologies Pvt Ltd
Shweta SinghShweta Singh
Support Engineer-IT at I-Telemetry Technologies Pvt. Ltd. (Subsidiary of Aaxis Nano Technologies Pvt. Ltd.)Support Engineer-IT at I-Telemetry Technologies Pvt. Ltd. (Subsidiary of Aaxis Nano Technologies Pvt. Ltd.)
Sowmya V BalasubramaniamSowmya V Balasubramaniam
Recruitment Manager at IBM India Private LimitedRecruitment Manager at IBM India Private Limited
Bharath Kumar SBharath Kumar S
Manager, Client Support Engineer at EpsilonManager, Client Support Engineer at Epsilon