Vignesh Saptarishi

AI systems engineering, machine learning, computational neuroscience

vignesh@traitful.ai | traitful.ai | saptaxis.dev
github.com/saptaxis | linkedin.com/in/vignesh-saptarishi

10+ years designing and engineering production AI systems across computer vision, NLP, generative models, agentic workflows, and ML platform architecture. Zero-to-one algorithms engineer on three products that reached production and revenue, built and led the ML and data organizations around them, technical expert on enterprise sales with Fortune 500 clients, and directly drove B2B revenue well into the tens of millions. Clients include FedEx, Macy’s, Levi’s, Tata Group, HDFC Bank, Diesel, Sundaram Finance, ThredUp and Dubai CommerCity. Doctoral research in computational neuroscience at Brandeis on an NIH fellowship, and a degree in electrical engineering.

Professional Experience

Oct 2024 - Present Founder, Traitful AI Labs

  • production RAG pipelines with optimized search engines for agent retrieval, external memory, scoped tool interfaces, demand-driven context assembly, and model and agent-topology routing per request
  • raw operational payloads of O(10k) tokens reduced to O(1k) in the model’s context, with first token under 5 seconds across multiple retrievals per turn
  • founded a custom AI engineering firm, shipping that core across freight documents and shipment intelligence, chat-to-dashboard over financial data, legal customer records, accountancy automation and sports auction bots; multi-tenant, end to end through IaC and Kubernetes deployment, as the sole engineer
  • desktop automation through accessibility APIs as the interface, with LLMs for planning and decisions
  • systems engineered for economic viability under real compute constraints, not big-tech budgets

Jan 2020 - Oct 2024 Director - Machine Learning, vue.ai / Mad Street Den

Built and led ML and data organizations. 100% code contributor across all projects alongside managerial responsibilities. Technology partnerships with Google Cloud, Meta, and Microsoft.

Enterprise AI Platform

The consolidation: every practice below, unified into one product.

  • one of the key architects of the company’s unified AI platform: four solution hubs over a shared data layer, an ML studio and DAG-based orchestration
  • shaped the platform’s entities and abstractions against every delivery pattern I had shipped: the algorithms, the data model, the workflows and the inference path
  • no-code surface for non-technical users to train models, compose workflows and map taxonomies themselves

Intelligent Documents Processing

  • zero-to-one founding algorithms engineer; led the ML and product org for domain-agnostic document processing with no separate product manager, the company’s fastest product to seven-figure ARR, now the platform’s Automation Hub
  • full stack from OCR to structured extraction: table and figure detection, custom embeddings, re-training loops; layout understanding as a hybrid of learned models (LayoutLM) and geometric heuristics (written up as a spatial layer)
  • one-shot extraction via hybrid text and image registration, a new document type onboarded from a single annotated sample with no training run
  • designed retrieval-augmented workflows before “RAG” had a name, inside a 4,000-token context window, matching sections against client taxonomies of thousands of attributes
  • non-LLM pipelines engineered for profitable unit economics at scale, deployed across government (Dubai CommerCity), BFSI (HDFC Bank, Sundaram Finance) and legal
  • algorithms: LLMs (in-context learning, RAG), document layout analysis, NER, classification, embeddings and search, OCR, summarization, computer vision, multimodal AI

Virtual Dressing Room

  • pioneered GAN-based virtual try-on, among the earliest production applications of GANs in this space
  • reframed generation as geometric deformation against a 512px generator ceiling and a 1600px-4K delivery requirement: regressing and upsampling morphing grids, then sampling from original images for photorealistic 4K output without hallucination artifacts
  • designed and trained core networks for garment transfer, lighting and shadow synthesis; made outfit visualization scale linearly rather than quadratically via one-time per-garment preparation
  • designed and operated a large-scale synthetic data pipeline: 3D modeling in Blender, garment physics in CLO3D/Marvelous Designer, automated rendering across hundreds of machines
  • engineered human-in-the-loop production systems: Photoshop-native mesh output with separate lighting and shadow layers, in-app inference plugins, and corrections captured back into training data, taking a retoucher from ~60 minutes a garment to 1-5 minutes
  • led a 6 member algorithms research team, and set training and technical direction for a 50+ person Photoshop production and data organization reporting into Product
  • algorithms: conditional GANs, spatial transformers, dense warp fields, shadow and lighting generation, segmentation (images and video), garment and body pose/parts detection

Feb 2017 - Jan 2020 Senior ML Engineer & Technical Lead, vue.ai / Mad Street Den

Automated Product Tagging for eCommerce

  • founding algorithms engineer; production computer vision for Fortune 500 retail and fashion enterprises including Macy’s, Levi’s, Diesel and ThredUp, now the platform’s Data Hub
  • built the tagging system on one internal fashion taxonomy with per-client naming maps: new-client delivery in a single working day, run by non-technical staff with ML on exception
  • small classification heads over frozen shared feature extractors serving hundreds of models from a shared fleet, with automated re-training loops and internal no-code model building
  • built the data collection organization: ~1 million labelled images in a few months with 10 people
  • algorithms: fine-grained hierarchical classification, object detection, garment and body segmentation, video-level scene understanding, cross-modal semantic search across image and text, generative product copy well before language models made this mainstream

Aug 2013 - Sep 2015 Graduate Researcher, Marder Lab, Brandeis University

  • simulated spatiotemporal models of neuronal homeostasis: coupled nonlinear conductance dynamics on spatially discretized, growing morphologies, and the feedback control that holds circuit behavior stable under perturbation, across multiple custom simulation environments
  • thesis: Homeostatic regulation of intrinsic neuronal conductances in morphologically growing neurons

Aug 2010 - Aug 2013 Research Associate, Waran Research Foundation, India

  • computational models and large-scale neural network simulators for the implication of cellular energetics on functional brain circuits

Independent Research & Open Source

2025 - Present Independent Research, Representation & Robustness

  • controlled study of how information channels shape learned controllers in a small parameterized physics testbed: with architecture and training held fixed, agents handed explicit physics learned smoother control and were less than half as robust under distribution shift as agents that had to infer it from consequences; read through behavioral signatures, not score
  • scad, session state and isolated execution for coding agents: indexes and archives every agent session on the machine from the agents’ own logs, and runs agents in containers that never touch the working tree (why the repo is the wrong boundary)
  • orglens, an organizational lens for AI agents: entity grammar as data over a docs tree, with a materialized topology that gives every session its organizational context on start; both ship agent-agnostic skills across model families

Education

Aug 2013 - Aug 2015 M.S. Neuroscience - Brandeis University, USA

  • admitted to the PhD program on a full NIH fellowship; graduate researcher in Prof. Eve Marder’s lab
  • graduate coursework in applied dynamical systems, mathematics and molecular biology; GPA 3.84/4.00
  • teaching assistant, undergraduate behavioral neuroscience with Prof. Don Katz

Sep 2009 - May 2013 B.Engg. Electrical Engineering - Anna University, India

  • coursework in control theory, numerical methods and signal processing; first class, GPA 8.0/10.0
  • thesis: Methods for Nonlinear Estimation using Kalman Filters to estimate state vectors and control DC motors

Fellowships & Awards

  • National Institute of Health - Neuroscience PhD Fellowship, Brandeis University, 2013-2015
  • Chairman (Devaki Muthiah) Endowment Award for Overall Best Student 2009 - 2013, Sri Venkateswara College of Engineering, Anna University