AI Lead with 10+ years in multi-agent AI solutions & system design
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
AI Lead specialized in architecting multi-agent AI solutions that automate complex enterprise workflows in the AEC, ConTech and Healthcare sectors. With 15+ years of system design and architecture experience in embedded firmware, kernel drivers, protocol design, and IoT platforms, I leverage deep expertise in building scalable, secure, and resilient systems to capture tacit knowledge, bridge domain gaps, and drive automation through applied AI.
Don Bosco Institute of Technology
B.E. · Electronics & Telecommunication
August 1, 2005 – June 30, 2009
AuxoAI
AI Lead
November 1, 2025 – Present
Mumbai, Maharashtra, India
Rently
Software Architect, Embedded
April 1, 2019 – March 1, 2024
Pune, Maharashtra, India
Phazr
Technical Lead, RnD
August 1, 2018 – April 1, 2019
Pune, Maharashtra, India
Greenvity
Senior Software Engineer
May 1, 2012 – August 1, 2018
Pune, Maharashtra, India
Acevin Solutions
Software Engineer
May 1, 2010 – May 1, 2012
Mumbai, Maharashtra, India
Automated-QTO
December 1, 2025 – April 1, 2026
Developed a highly robust Quantity Take-off system for complex civil water works drawings, successfully handling extreme variations in drawing quality, including noisy PDFs with heavy overlapping lines (utilities, road markings, existing pipes), symbols, annotations, and callouts. Processed challenging multi-contractor project PDFs containing drawings from various disciplines, scales, layouts, legends, and keynotes with high generalization and accuracy. Adopted a contrarian AI Agent-first approach (instead of traditional CV-first methods), leveraging the rapid advancements in State-of-the-Art (SOTA) vision reasoning models to achieve superior performance on highly variable real-world drawings. Designed a novel Filter - Generate - Ground pipeline specifically optimized for balancing recall vs. precision: Filter Step: Pure LLM-driven filtering (no heuristics) that dynamically creates filters based on PDF metadata such as line thickness, tags, and attributes to eliminate obvious noise and irrelevant elements. Generate Step: Innovative use of an image generation model (nano-banana) to intelligently isolate target pipes from the filter step, followed by accurate vectorization using potrace. Ground Step: Applied advanced vector clustering on directional pipe segments using dynamic X/Y tolerances to eliminate false positives and ensure high precision. Evaluated open-source Vision-Language Models (VLMs) such as the Qwen Image Layered model for improved pipe approximation capabilities within the creator-critique workflow. Orchestrated a sophisticated multi-agent system with ~12 specialized agents responsible for context distillation, textual and visual drawing understanding, validation, and final quantity computation. Implemented a comprehensive validation agent that collated spatial / textual / visual agent artifacts and called false positives, identified missing symbols, inconsistencies using techniques focused on presence, fidelity, spatial consistency, reference subjects, and plan description. Achieved an excellent balance between generalization (handling wildly different drawing styles and quality) and accuracy (reliable quantities even in extremely noisy conditions). Successfully tackled one of the most difficult problems in AEC automation by breaking it down into smaller, solvable sub-problems through intelligent agent orchestration.
Headconn
August 1, 2025 – August 1, 2025
An AI agent that can create a composite picture using a short mash-up scene pitch of two pop-culture references. It understands main characters and objects from popular culture including movies and books. It uses image tools like crop, remove background, rotate, re-size, shear and composite tool to create a mash-up by superimposing the objects and characters over each other to create a mash-up. It uses grok-4 instruction following and structured outputs to call tool functions. In v2, it uses muli-agent. Imagine agent for creating a creative script and searching images with proper characters/objects in the foreground. Reflect agent to short list the best matched photos. Composite agent for creating the composite image.
View ProjectRestaurant-FL
March 1, 2025 – July 1, 2025
A privacy preserving food intellect and a prediction model for customer insight, local and global trends. Uses natural language to extract features and heterogenous graph transformers to capture high-order interactions between restaurants, customers and locality. Generates customer description from chat summary and stores features and interactions as nodes and edges in the knowledge graph. Uses a dataset from kaggle for restaurants, but creates its own customer data flywheel. Trains local ML model with customer data and averages trained weights across all restaurants for a global model. Integrates gemma3:4b for chatbot (via ollama), Flowerai for federated learning, HGT (Heterogeneous Graph Transformer) for ML model with custom downstream layers, Graphiti for knowledge graph, Status-im for p2p messaging and Tendermint for consensus.
View ProjectDecentralized AI Bootcamp
Encode Club
June 1, 2026 – Present
The Mathematics of Cryptography
Dr James Grime
June 1, 2026 – Present
Cultural Fit Analysis
The candidate's project portfolio showcases a diverse range of applications, from AEC automation to privacy-preserving federated learning and creative AI agents, indicating adaptability and a broad interest in AI applications. Their experience spans various industries (AEC, Healthcare, Smart Home) and roles (AI Lead, Software Architect, Technical Lead), suggesting a versatile professional who can thrive in different organizational cultures. The focus on SOTA models and continuous learning aligns well with an innovative, growth-oriented culture.
Soft Skills & Operational Fit
The candidate demonstrates strong problem-solving abilities, intellectual curiosity, and a detail-oriented approach. Their experience in leading discovery sessions and technical discussions with clients indicates strong communication and stakeholder management skills. The 'builder-leader' and 'player-coach' attributes suggest a hands-on leadership style suitable for fast-paced startup environments. The emphasis on capturing tacit knowledge and bridging domain gaps highlights a strategic operational fit for driving enterprise AI automation.