Vik ████ (hidden, sign in to see it)

Senior AI Engineer

  • Enterprise Agentic Systems & AI Platforms
  • Regulated Medical Devices
  • Los Angeles, CA (open to relocation)
  • phone hidden (hidden, sign in to see it)
  • email hidden (hidden, sign in to see it)
  • github.com/engineer6080
  • U.S. permanent resident (no sponsorship required)

Summary

Senior AI engineer with 4+ years shipping validated GenAI, ML and computer-vision systems in a regulated medical-device environment. I started in embedded systems and have worked my way up the stack to multi-agent orchestration. Recently enjoyed tackling the interdisciplinary challenges of developing an enterprise Agent Catalog: navigating architecture, data and governance concerns alongside effectively abstracting harness operations. I’ve learned to build agents people can trust through practices such as shaping the tooling around observations, maintaining shared understanding and providing transparency of outcomes & capabilities, to ultimately earn and sustain adoption.

Global medical-diagnostics company (hidden, sign in to see it)

Remote (hidden, sign in to see it)Jun 2022 – Present

  1. Senior AI/Data Scientist, Enterprise Data & AIAI Council MemberOct 2025 – Present
  2. AI/Data Scientist, Digital TransformationOct 2022 – Sep 2025
  3. Data Science Intern, Advanced AnalyticsJun 2022 – Oct 2022

Enterprise Agent Platform & AI Governance

  • Agent CatalogMCP + MCP Apps, A2A, AG-UIArchitect & Platform LeadSecured CTO sponsorship at an executive kaizen to build an internal, governed, model-agnostic agent platform on open protocols. Specialist agents are organized by business domain, each aligned to SME process and knowledge, with a curated tool surface and shared harness for internal operations.
  • SecurityDesigned the security architecture and data connections and cleared technical/governance approval chains. Enforced RBAC at the data layer through end-to-end OAuth propagation; eliminated standing direct access by routing data sync through elevated tools on a stateless MCP server.
  • QualityLayered agent test suite: foundational (protocol, state schema), deterministic input/output tests on the tool surface, and scenario tests that verify chained tool responses; user sentiment and feedback loop fed by reported friction and interaction failures; every capability validated and documented transparently.
  • OpsDevelopment lifecycle, change management, environment and release control, auto-scaling on AgentCore, monitoring, identity/RBAC, and cost, usage, feedback and ROI reporting.
  • Impact5 agents shipped in 6 months across Manufacturing (OBQ, floor troubleshooting (hidden, sign in to see it), Non-Conformance with an investigative orchestrator), IT (OBIEE report discovery) and Sales & Marketing (IFU); 2 in validated production, 100+ users overall.
  • Enterprise LLM GatewayAzure AI Foundry (incl. Anthropic)Built & OperateIntroduced AI usage tracking, spend reporting and granular oversight, serving hundreds of developers and client apps across OpCos within the tenant, with allocation by role tier and per-client quotas (tokens/day, TPM/RPM). Verified by reconciling calculated token costs with billed invoices.
  • Governance, Evangelism & AI CouncilOne of the first custom-agent use cases through the newly formed AI governance process: surfaced and closed remaining process gaps (intake logistics, GPO risk assessment/tier, TRB/ARB review), co-authored the architectural standard and worked with Cloud Ops to extend the Jenkins deployment pipeline to enable more protocols: AgentCore runtime option (MCP/Apps, A2A, AG-UI).
  • Socialized agentic capabilities across the org: live demos to senior business-function leaders using a catalog agent from their own domain, created self-driven WebMCP demos, covering customization options and the underlying protocols and frameworks; populated the use-case funnel with informed ideas and raised org-wide AI awareness.
  • Databricks AI PlatformAzureAdminSet up the AI team's Databricks platform, brought it inside VNet (hub-spoke, private subnets) with egress to VPN-protected services; Unity Catalog; ~50 users build and collaborate securely.

Agents & Applied AI

  • OBQ (Out-of-Box Quality) AgentManufacturingProductionSupports installation-failure investigations by automating KPI calculation and monitoring, classification for Pareto analysis and topic-triggered follow-up with field service engineers (FSEs); ~18–30 hrs/week saved at one site (hidden, sign in to see it); expansion to two more sites (hidden, sign in to see it) and additional product lines planned.
  • Floor Troubleshooting (hidden, sign in to see it) AgentManufacturingProductionUnified fragmented diagnostics and assembly data sources into a single multimodal troubleshooting lookup for the manufacturing floor.
  • IFU (Instructions for Use) AgentSales & MarketingPilotBuilt grounded Instructions-for-Use search over a 15K-document controlled corpus for sales, RFP and support teams; 98% accuracy in phase-1 validation (diverse 70 question set) vs. ~30% for off-the-shelf AI (business baseline), ~80% thumbs-up; business-estimated ~21.8K hrs/year; phase 2 rolling out to 25 sales reps.
  • Complaint TrendingQualityBuilt issue trending over 2.5M+ complaints across 5 product lines for Complaint Handling Unit; GPT-4.1 derives persistent issue groups and classification rules, standardizes and translates new cases, and a reasoning model flags anomalies and potential systemic issues, each backed by counts and case numbers.

On-device vision inspection

  • Vision-inspection apps (hidden, sign in to see it): Diffuser & FramesIncoming inspection (hidden, sign in to see it)iOSNon-product software validationOn-deviceDigitized incoming inspection on standard iOS devices: validated and productionized two inspection apps combining OpenCV with on-device classification models to guide and log inspections and highlight areas of concern.
  • DiffuserUses built-in iOS sensors (IMU) to align, measure light diffusion of glass surfaces and verify conformance, replacing a manual inspection whose defect escapes had been tied to 2 CAPAs and 2 stop-shipments. Established the non-product software validation (NPSV) precedent and Apple enterprise signing/distribution path for iOS apps.
  • FramesReal-time AR inspection of aluminum chassis: a marker-free CoreML landmark model + OpenCV that superimposes a template to classify surface features, highlighting any that deviate from the expected class live in the inspector's view, fully on-device. Owned development, validation and distribution for both apps.

More projects

  • Supplier order reconciliationExtracts and matches multilingual supplier documents (~50 vendors) against open Oracle POs using vendor-specific document rules, partial-match rescoring and priority scoring; raised the open-PO line match rate from ~25% (manual) to ~80%, surfaced in Power BI; 15K+ documents processed; live at a European site (hidden, sign in to see it).
  • Oracle AR error resolutionTook a winning hackathon prototype to production in 3 regions; triages interface errors and proposes fixes for human approval, cutting blocked-sales-order review time ~90%.
  • GenAI Power BI visualA custom Power BI visual that embeds catalog agents over A2A/AG-UI via an MCP endpoint behind APIM; started as a hackathon entry, now in production.
  • DFMEA engine prototypeLLM-assisted engine (hand-built tool routing and state over LangChain, pre-MCP) that cross-references historical DFMEAs, regulatory citations and complaint trends to draft failure-mode entries with full source lineage.
  • Ticket-routing classifierTrained and deployed a TF-IDF + BERT classifier (TensorFlow) to production for group, department and subgroup routing; later superseded by an LLM rules engine.
  • Developer enablementHackathon program advisor and mentor; lead of org-wide GenAI developer workshops.

Cal Poly Pomona

Pomona, CASep 2020 – Jun 2022

  1. Graduate Research Assistant, Computer ScienceAerospace Dept.Sep 2020 – Jun 2022

Precision-agriculture robotics

  • Built the autonomy stack for a tracked UGV (SuperDroid Robots HD2-S Mastiff) in funded precision-agriculture research: reverse-engineered the SuperDroid Robotics drive-control protocol, adapted Xsens IMU driver for Jetson Nano, ran RTK GNSS (Emlid) waypoint navigation and onboard classification with sensor fusion through 6–7-hour field runs in rugged, high-temperature, offline farm environments.
  • Coordinated multispectral, hyperspectral and LiDAR UAV flights with the UGV and processed terabytes of imagery into geo-rectified orthomosaics and point clouds; responsible for field-to-HPC data transmission for the Cal Poly / NASA / USRA fire-detection collaboration.
  • In the field: DJI Matrice UAVs carrying MicaSense Altum, Headwall hyperspectral, Applanix GNSS/INS and Phoenix LiDAR payloads; real-time segmentation deployed on-site and a live orthorectified tile map served from the UGV stream.

Sinsat, Inc.

now Katsh Digital ID

Los Angeles, CANov 2019 – Jan 2021

  1. ML Engineer, Computer Vision Leadpart-time from Sep 2020Nov 2019 – Jan 2021
  • Built dorsal-hand biometric feature extraction (OpenCV + CNN) and an on-iPhone real-time segmentation engine for anonymized identity verification, plus the AWS backend and identification API; developed working iOS prototype and company website.

Earlier projects

Embedded and robotics roots

  • NATCAR autonomous race carUC DavisTeam of 42017–18A 1/12-scale car that steers itself around a track, on the team's own PCB: VNH5019A H-bridge motor driver, three separate ground planes to isolate high motor currents, and headers for the camera microcontroller, servo and Bluetooth. An OpenMV M7 reads the line (blob detection, then linear regression) and feeds PID steering, with Hall-effect speed control and a Bluetooth hard stop.
  • Autonomous RL carUdacity capstoneCompared lane tracking across reinforcement learning (DQN), a PID control scheme and an NVIDIA-style end-to-end network, on a camera car with an Arduino Nano motor controller.
  • Deep RL quadcopterTrained a simulated quadcopter to hover and follow a trajectory with Deep Deterministic Policy Gradient.
  • Systems workMulti-threaded C++ linear algebra library; operating-systems projects (FAT file system, thread library, kernel, shell); OV2640 camera on a CC3200.

Skills

  • AI & AgentsMCP + MCP Apps, A2A, AG-UI; PydanticAI, LangChain, CopilotKit; AzureOpenAI, Anthropic, VertexAI, Bedrock, Ollama (local/offline), Jev/Decision-1; multi-agent orchestration, tool calling (frontend/backend), evaluation harnesses, WebMCP/Playwright, graph, guardrails, HITL, OpenTelemetry
  • RAGMultimodal; embeddings (text-embedding-3-large, Qwen3-VL, Titan, Arctic); pgvector, Databricks Vector Search, Cortex Search, ChromaDB; document parsing/OCR (Azure Document Intelligence, DeepSeek-OCR, GLM), structure-aware chunking; metadata capture; knowledge graphs (Apache AGE, Memgraph)
  • Data & CloudDatabricks (Unity Catalog, Spark, MLflow), Snowflake, Postgres, Power BI, Oracle EBS-to-Snowflake ETL; Azure (APIM, Foundry, Container Apps, VNet), AWS (Bedrock, AgentCore, ECS, RDS, PrivateLink), GCP (Vertex AI); Docker, Jenkins CI/CD; Salesforce/ServiceMax, Oracle EBS/OBIEE, SmartSolve QMS
  • Identity & GovernanceEntra ID, OAuth 2.0 (OBO, identity federation, client credentials, SSO propagation), source-preserving RBAC; AI governance, non-product software validation (21 CFR 820 / QMSR, ISO 13485), TRB/ARB, CAPA, D-FMEA, IP/PII handling, zero data retention
  • Languages & MLPython, TypeScript, SQL, C/C++, Objective-C, Swift, CUDA; PyTorch, TensorFlow, Transformers, CNNs (segmentation, object detection), Core ML, OpenCV, time-series forecasting
  • Embedded & RoboticsC/C++ firmware; Jetson Nano, Raspberry Pi, STM32/ESP32/Arduino; UART/USB serial and custom packet protocols; IMU/GNSS drivers; sensor fusion, RTK GNSS, LiDAR, multispectral/hyperspectral sensing

Education

  • Cal Poly PomonaM.S. Computer Science, coursework and funded research (left to join current employer (hidden, sign in to see it) full-time)2020–22
  • UdacityMachine Learning Engineer Nanodegree, Autonomous RL Car Project2019
  • UC DavisB.S. Computer Engineering2018

Publications & Recognition

Proc. SPIE 12114 (2022): 4 conference papers; first author on 3.

  • UAV–UGV collaboration for weed classification and detection (hidden, sign in to see it)First author
  • Real-time plant-health assessment from UAV RGB imagery (hidden, sign in to see it)First author
  • Correlating UAV remote sensing with proximal sensor data (hidden, sign in to see it)First author
  • Hybrid bermudagrass health via remote sensing (hidden, sign in to see it)Co-author
  • HackathonWon 2 consecutive internal hackathons at current employer (hidden, sign in to see it); awarded $100K internal funding to build a real-time compliance-feedback engine inside ServiceMax (custom MAS)
  • PosterVision-inspection apps (hidden, sign in to see it), medical extended-reality summit (2024) (hidden, sign in to see it)
  • FeaturedUAV-based remote sensing for precision agriculture and bringing AI to farms (hidden, sign in to see it), CSU STEM-NET story (2020)
  • ConferencesDatabricks Data + AI Summit; AWE 2024 (Augmented World Expo)