P Asik Kani

Systems Engineer & Agentic AI Architect

Architecting low-latency autonomous pipelines, local LLM fine-tuning, and hardware-to-OS system integrations.

I bridge the gap between heavy AI model reasoning and bare-metal OS control. Currently operating as an ML Engineer Intern at Aathma Pranavaayu, a Core AI/ML Department Member at AWS Cloud Club VITC, and an active contributor within the open-source ecosystem.

Out of 20+ competitive national hackathons, these five builds survived strict 24-to-48-hour timelines and intense judging panels to secure overall podium positions:

Aura-OS — Autonomous Android Executive Layer

• The Core: Built a zero-hop edge framework enabling a Gemini 2.0 Flash ReAct loop to autonomously control third-party mobile applications without hardcoded paths.
• System Mechanics: Leveraged AccessibilityService and a custom SemanticFlattener to convert dense UI layouts into tokenized JSON, dispatching execution via the GestureDescription API and Jaro-Winkler string tracking.
• Race-Condition Mitigation: Solved real-time streaming constraints across bidirectional JSON/Audio streams using a thread-safe ConcurrentHashMap combined with a deterministic 3x 800ms retry draw-cycle strategy.

AgentChain — Transparent Trustless Web3 Trading

• The Core: Architected an autonomous crypto trading agent running on the Model Context Protocol (MCP) that moves decisions entirely on-chain.
• System Mechanics: Decoupled multi-variable trade logic into explainable decision trees. Every execution is logged immutably via smart contracts, generating a transparent ledger audit track for DAO governance.

VoicePulse — Conversational Diagnostic Console

• The Core: A voice-first analytics engine running conversational AI interviewers that dynamically deep-probe open-ended verbal responses.
• System Mechanics: Engineered a real-time transcript processing engine featuring dynamic topic segmentation and context-aware sentiment analysis to spit out priority action matrices for systems operators.

ChronoPet — Native Windows Active Focus Companion

• The Core: A low-friction background desktop organizer that parses task state via semantic memory loops.
• System Mechanics: Monitors foreground processes directly via native Win32 APIs paired with local semantic memory handling via ChromaDB. Task ingestion runs via an isolated Docker container hosting a WhatsApp message parsing microservice via LiteLLM.

SignalWorks — Zero-UI Predictive Market Engine

• The Core: Built a network decision infrastructure giving 10 Million+ offline street vendors live geospatial market analytics without internet access or smartphones.
• System Mechanics: Combines Amazon Bedrock (Claude 3 reasoning) and SageMaker pricing trends. Delivers real-time location heatmaps and localized data via 2G SMS and automated Interactive Voice Response (IVR) phone loops. [Shortlisted for AWS AI for Bharat]

TerraGuard — Distributed Seismic Mesh Network

• The Core: An end-to-end telemetry system designed to flag and intercept illegal river sand mining operations using remote hardware nodes.
• System Mechanics: Deployed low-power ESP32 microcontrollers running local signal-processing algorithms to isolate machinery vibration frequencies on the edge, routing clean alerts to an Orange Pi 5 gateway broker.