Amoghraj Lakkanagavi

As an AI Platform Architect, I thrive on the autonomy to design and implement robust, compliant solutions, driven by a strong sense of accountability and ambition. My idealistic vision guides me in crafting innovative AI platforms, ensuring precision and adherence to the highest standards.

Experience

O

AI Platform Architect

Ojas · Present

Not yet confirmed

• Architected Ojas, a HIPAA-compliant multi-tenant API platform exposing specialised medical AI via a single developer-facing API  with RBAC, credit-based billing, real-time SSE streaming, webhooks, PHI-safe data handling, and GPU-accelerated model serving on GCP Kubernetes.
• Designed per-request audit logging and tenant data isolation meeting HIPAA technical safeguard requirements  enabling the platform to be used by third-party healthcare developers without compliance risk.
• Introduced spec-driven development across the team  a custom SDD workflow where PMs defined tasks as structured specs and engineers implemented against them with Claude Code-assisted coding; shipped the full Ojas platform in under 2 weeks.

M

On-device Health Monitoring Lead Engineer

Maí · Present

Not yet confirmed

• Built continuous vitals monitoring using an on-device SLM  real-time health signal analysis with full offline operation and a context-aware alert engine that surfaces clinically significant patterns while suppressing noise.
• Integrated smart band over BLE; architected a CDC-compatible [band  device  cloud] sync pattern  decoupling the app from cloud dependency for core monitoring with reliable eventual consistency on reconnect.

O

AI Research Engineer

On-device AI Research & POCS · Present

Not yet confirmed

• Delivered end-to-end speech-to-speech on-device: ASR  LLM reasoning  TTS  fully local, no cloud round-trip; demonstrated production-grade conversational AI latency on constrained hardware.
• Built runtime model swapping (vision/text) on-device; implemented on-device function calling pipelines; developed semantic router for intent-to-model dispatch via embedding similarity; validated Gemma models for offline latency-critical deployment.
• Benchmarked full-system TTFT on-device: ASR + SLM function calling achieves 1-2s; ASR + SLM + VLM pipeline achieves 3-5s  all running on-device with zero cloud dependency.

Software Developer 2 (AI-CV)

AjnaLens · Feb 2023 – Present

Not yet confirmed

• ▸ Agentic AI Platform — Consumer Wearables
• Agentic AI platform for a consumer smart glasses product — multi-step task execution, persistent memory, and MCP integration, deployed on Kubernetes sustaining 10× traffic spikes at p99 under 500ms.
• Built agent memory for long-horizon continuity and session-aware personalisation; full observability across trace logging, latency dashboards, and error attribution.
• ▸ On-device AI — Smart Glasses NPU
• Ported a production ASR model to a wearable NPU — optimised for constrained memory and compute; fully offline, zero cloud round-trip.
• Built a wake word pipeline: data generation, model training, and NPU inference runtime — enabling hands-free glasses activation on-device.
• ▸ Clinical AI — Government Hospital Partnership
• Led AI engineering under a tripartite clinical partnership — full ownership of two medical AI products (dermatology and health screening) from scoping through production.
• Fine-tuned vision models on clinical datasets from a government hospital; built a PHI-compliant pipeline with encryption, access controls, and audit trails for cross-institution data exchange.
• Implemented a continuous retraining loop with quality-gated evaluation before any version hit production.
• ▸ HIPAA-compliant Healthcare AI Developer Platform
• Architected a HIPAA-compliant multi-tenant API platform exposing medical AI to third-party developers — RBAC, credit billing, SSE streaming, PHI-safe data handling, and GPU-accelerated serving on Kubernetes.
• Shipped the full platform in under 2 weeks via spec-driven workflow with AI-assisted engineering.
• ▸ On-device AI Research
• Delivered end-to-end speech-to-speech on-device (ASR → LLM → TTS) — production-grade latency on constrained hardware, zero cloud dependency.
• Built runtime model swapping, on-device function calling, and a semantic router for intent-to-model dispatch.
• Benchmarked TTFT: 1–2s for ASR + SLM; 3–5s for ASR + SLM + VLM — entirely on-device.

A

Lead AI/CV Engineer (SDE II)

Ajnalens · Feb 2022 – Present

Not yet confirmed

• Architected and productionised an agentic AI platform  multi-step autonomous task execution with scalable backend sustaining high-throughput LLM inference at p99 response times under 500ms; deployed on GKE with horizontal pod autoscaling, sustaining 10 traffic spikes without SLA degradation.
• Built a persistent agent memory system enabling long-horizon task continuity, session-aware personalisation, and cross-conversation state eliminating repeated context reconstruction overhead.
• Integrated MCP (Model Context Protocol) and built developer quality-of-life tooling enabling teams to extend agent capabilities without modifying core infrastructure; established full observability  trace logging, latency dashboards, error attribution.
• Ported Whisper ASR to Qualcomm AR1 Gen 1 NPU  optimised the model for the wearable's constrained memory and compute envelope while preserving transcription accuracy; no cloud round-trip required.
• Built a custom wake word pipeline end-to-end: synthetic and real data generation, model architecture, on-device classifier training, and NPU inference runtime enabling fully offline hands-free glasses activation.

A

Clinical AI Lead (Indus Derma & Aarogyam)

Ajnalens (in partnership with AIIMS  Google) · Feb 2022 – Present

Not yet confirmed

• Led the AI engineering team under a tripartite partnership with AIIMS and Google, taking full technical ownership from problem definition through production deployment of two medical AI products: Indus Derma (dermatological diagnosis) and Aarogyam (health screening).
• Fine-tuned vision models on the SigLIP architecture using Indian clinical datasets from AIIMS, adapting representations to Indian dermatological presentations underrepresented in standard benchmarks  achieving clinically validated accuracy.
• Built a PHI-compliant secure data pipeline for clinical imaging handover  encryption, access controls, audit trails, and compliance for sensitive medical data transfer between a government hospital and a private organisation.
• Implemented a Vertex AI continuous retraining loop: clinician-labeled data feeds back automatically into training, with model evaluation gates enforcing quality thresholds before any version reaches production.

Full Stack Developer

AjnaLens · Feb 2022 – Feb 2023

Not yet confirmed

• Platform Revamps: Spearheaded major overhauls of collaborative web platforms, taking full ownership of interactive web applications with real-time content editing and cross-functional workflows.
• End-to-End Development: Led development of large-scale systems from scratch, including frontend architecture, backend APIs, and database design—ensuring high performance, scalability, and user-centric design.
• Blockchain Integration: Engineered and deployed Ethereum-based smart contracts (ERC20) on Layer 2 solutions like Polygon, handling token deployment, feature planning, and secure integrations.
• NFT & Web3 Applications: Researched, developed, and integrated NFT trading features including tokenization, swapping, and on-chain listing—interfacing directly with third-party vendors and ensuring seamless UX.
• Skills & Tools: FastAPI, React, WebVR, Solidity, AWS, PostgreSQL, RESTful APIs, Smart Contracts, System Design, Information Architecture, Tokenomics, Blockchain, and CI/CD pipelines.

T

Flutter Developer

TRiDE Mobility - EV Rides · Oct 2021 – Feb 2022

Not yet confirmed

• • EV Vehicle Marketplace: Ideated, designed, and developed features for an EV vehicle marketplace, including backend API development and cross-platform UI components in Flutter.

Freelance Developer

Freelance · Sep 2018 – Oct 2021

Not yet confirmed