AI / GenAI and Mobile Engineer · 4+ years of mobile engineering · Building production systems
I’m a software engineer with 4+ years of experience shipping consumer products at scale. I’m now focused on Applied AI / GenAI, building systems around retrieval, fine-tuning, real-time voice, agents, and code intelligence.
Currently working as a Mobile Engineer at Anko GCC, building for Kmart & Target Australia.
- Retail platform serving 3.74M active users
- WebView memory: 2GB → 400MB
- Token clearing: 2,000ms → 10ms
- PLP first load: 6–9s → 131ms
- PDP logged-in: 3,054ms → 231ms
- Next-Gen Home Screen: 38.9M sessions
- Shoppable UGC: 17M events · 923K users · 24% conversion among interactors
- Connectivity recovery improved by 2–7s
- Ran engineering/code-quality sessions for 10–12 engineers
Earlier, I worked across startups and product teams building consumer applications, including social investing and digital dining platforms.
Real-time AI phone-call platform with streaming STT → LLM → TTS pipelines.
- Cascaded streaming pipeline: Twilio Media Streams → Deepgram STT → Grok/Kimi → Cartesia/ElevenLabs TTS
- Barge-in & turn-taking: endpointing, cancellable speech tasks, and Twilio clear events
- Latency engineering: sentence-level LLM → TTS streaming with per-turn avg/p95 latency tracking
- Real phone calls: 8kHz μ-law audio over WebSockets
- Configurable workflows: loan recovery, EMI, banking, and sales
- Structured captures: promise-to-pay, lead qualification, escalation
- Multilingual: English, Hindi, and Hinglish with language-aware STT/TTS
Stack: Python FastAPI WebSockets Twilio Deepgram Grok Cartesia ElevenLabs SQLite Next.js
Point it at a repository → understand the codebase, dependencies, risks, execution flows, and changes.
GitHub · Website · Demo Video
- Corrective RAG: parent-child chunking, semantic search, symbol lookup, graph expansion, and query rewriting
- AI code review: repository-aware diff review, language-specific rubrics, blast-radius analysis, and test-verified fix suggestions
- MCP server: 39 tools for VS Code Copilot, Claude Code, and Cursor
- Graph intelligence: DuckDB + igraph for PageRank, blast radius, cycle detection, and dependency analysis
- Agentic workflows: LangGraph chat, human-in-the-loop gates, and deep-research fan-out
- Code parsing: tree-sitter across 15+ languages
- Incremental reindexing: only changed files are re-processed
- Multi-provider LLM layer: 7 providers with retry and fallback
- Semantic code search: vector retrieval over repository code
- Execution flow: entry points and module/file dependency chains
Stack: Python LangChain LangGraph ChromaDB DuckDB igraph tree-sitter FastAPI Next.js Docker
| Interface | Description |
|---|---|
| MCP Server | 39 tools for AI coding agents |
| REST API | Repository analysis, chat, review, indexing, graph intelligence |
| Web UI | Visual repository exploration, diagrams, blast radius, reviews |
Fine-tuned small language model + deterministic business logic for customer support.
- Fine-tuned Qwen 2.5 0.5B Instruct with LoRA
- Used PEFT + TRL
- Trained on 385 customer-support conversations
- Covered orders, returns, and refunds
- Hybrid inference: LLM extracts structured intent while Python performs catalog/order lookups
- Keeps business-critical responses grounded in actual application data
Stack: Python FastAPI PyTorch Transformers PEFT TRL Qwen 2.5
AI-powered trip planning application shipped across web, Android, and iOS.
Web · Android · iOS · Sample Trip
Rebuilt Flutter's ListView from scratch with virtualized rendering supporting fixed and dynamic heights.
17 iOS applications built while learning Swift and UIKit.
A local-first notes application.
RAG · LangChain · LangGraph · MCP · Embeddings · ChromaDB · PEFT / LoRA · TRL · PyTorch · Prompt Engineering · Function Calling · Real-time Voice · LLM Evaluation · Langfuse
Python · FastAPI · WebSockets · REST · Docker · PostgreSQL · DuckDB · SQLite · CI/CD
Flutter · Dart · BLoC · SwiftUI · Firebase · Platform Channels
Python · Dart · Swift
I care about measurable engineering outcomes, not just shipping features.
| Area | Result |
|---|---|
| WebView memory | 2GB → 400MB |
| Token clearing | 2,000ms → 10ms |
| PLP first load | 6–9s → 131ms |
| PDP logged-in | 3,054ms → 231ms |
| Home Screen | 38.9M sessions |
| Shoppable UGC | 17M events · 923K users |
| Connectivity recovery | 2–7s faster |
| Engineering enablement | 10–12 engineers |
- GitHub: https://github.com/gupta29470
- LinkedIn: https://www.linkedin.com/in/aakash98gupta/
- Email: aa.1998.gupta@gmail.com
Building production-grade Applied AI / GenAI systems and looking for opportunities where I can combine my software engineering background with AI systems engineering.
Interested in: RAG · Agents · LLM applications · Voice AI · AI infrastructure · Model fine-tuning · Developer tools