Full-Stack Developer · AI Systems

I build reliable systems that people depend on every day. Full-stack Laravel, React and Next.js for logistics and freight forwarding, plus open-source agentic AI: multi-agent systems, RAG pipelines and FastAPI.

@yashoncode

2,061 contributions all time

Private contributions included, counted without repository details.

01 · About

Full-stack systems, now with agents.

Software Engineer at Stockarea building scalable full-stack applications with Laravel, React, and Next.js. I develop software that streamlines logistics and freight forwarding operations, plus AI-powered automation workflows that improve business processes and operational efficiency. Deeply into Generative AI, and actively building open-source AI projects with Python and FastAPI: LLM applications, multi-agent systems, RAG pipelines, AI document analysis, and AI-powered SQL query generation.

Agentic AI Systems

Multi-agent systems, RAG pipelines, and agentic knowledge bases in Python and FastAPI: LangGraph, LangChain, MCP, pgvector, Langfuse.

Backend Ownership

Laravel REST APIs, CRM & ERP modules designed, built, and run end to end, often as the sole engineer.

Full-Stack Delivery

Laravel, React, and Next.js apps for logistics and freight forwarding, design to deploy via CI/CD, Docker, and Coolify.

AI Workflow Automation

n8n, LLM-driven document handling, and custom tools automating daily operations for an entire company.

02 · Experience

My journey so far.

Software Engineer L1 → L2

@ Stockarea

Oct 2024 - Present · L2 since Apr 2026

  • 3 GB → 200 MB

    Diagnosed and fixed a major report-generation bottleneck: memory cut from ~3 GB to ~200 MB, response time from ~5 min to ~5 sec, by restructuring how large datasets are queried and processed.

  • 493 ms → 0.8 ms

    Replaced substring LIKE scans with a Manticore full-text index, synced from MySQL by a scheduled job so indexed models pay nothing on write. Benchmarked at 1M rows: 493 ms to 0.8 ms, and flat as the table grows rather than linear in table bytes. Adds typo-tolerant matching LIKE cannot do at all.

  • Company-wide automation

    Built and maintain the workflow-automation tools every team runs on, streamlining daily operations across the company.

  • One admin API, four domains

    Build and own the Laravel admin API powering core logistics workflows across CRM, ERP, positions, and asset management.

  • CRM built, end to end

    Designed and built the company's CRM modules and workflows end to end, with AI-powered automation integrated into the flows.

  • Asset management backend

    Designed and built the positions & asset-management backend end to end as the sole engineer, covering data structuring, architecture, and implementation.

  • Full-stack with Next.js & React

    Developing internal web applications, extending ownership from backend APIs to full-stack delivery.

  • AI automation in production

    Build AI-powered automation workflows into logistics and CRM processes, with LLM-driven document and data handling that removes manual steps and speeds up operations.

  • Finance module & Zoho Books

    Built and maintain the finance module: invoices, purchase orders, and bills synced with the Zoho Books API over webhooks, with automated profit and revenue reporting for the finance team, plus vendor bill-payment and expense tracking.

  • AI voice calls in production

    Three outreach channels for the sales team on n8n with open-source LLMs: email, WhatsApp reminders, and AI voice calls handling 100+ calls a week, with constrained prompts, response validation, failure logging, and automatic retries. Follow-up work that filled a rep's day now takes 5 to 15 minutes.

  • Own the whole pipeline

    End-to-end delivery: design, testing, debugging, and deployment via GitHub CI/CD, Docker, and Coolify.

Solutions Architect, Automation

@ Stockarea

Aug 2023 - Oct 2024

  • Automation foundation

    Designed and built the base of Stockarea's internal automation, translating business operations into reliable backend workflows.

  • Systems integration

    Backend integrations with Laravel and Google APIs connecting disparate internal tools.

  • Data & reporting

    Reporting and data-management with Metabase, MongoDB, and SQL to support operational decision-making.

  • Early tooling

    Workflow tools in Google Apps Script and AppSheet that became part of daily operations.

Founder & Software Lead

@ Flare Innovations

Feb 2022 - Jun 2023

  • Founded the company

    Founded and led development of IoT-based monitoring systems for aquaculture and hydroponics environments.

  • Real-time sensor data

    Python applications processing and analyzing live sensor data from hardware devices.

  • Hardware → cloud

    Integrated hardware sensors with cloud systems for remote, real-time monitoring.

  • Led a 10-person team

    Led a 10-person team through product prototyping, deployment, and testing.

  • Incubated and deployed

    Incubated at ICAR-CIBA (Central Institute of Brackishwater Aquaculture) and Annamalai University, with a pilot system deployed at CIBA. Wound down after graduation to join Stockarea.

03 · Projects

Open-source AI, built in the open.

Reyaak
source

An autonomous agent that runs on the phone itself, with a multi-armed bandit picking which of 37 LLM providers answers each turn.

  • The router scores 37 free-tier providers every turn. Beta-posterior reliability with Thompson sampling, so exploration is proportional to uncertainty rather than a fixed epsilon, with headroom and rate-limit guardrails as always-on multipliers.
  • Memory and skills are separate concerns: facts the agent can search, and procedures it writes for itself. Curation runs on inactivity, archives instead of deleting, and refuses to touch a shipped skill or one you edited.
  • Native Kotlin, not a WebView wrapped around a bundled runtime. The :ui module has no platform source set, so an iOS host could supply its own fonts without :ui gaining an expect declaration.
KotlinJetpack ComposeRoomCoroutinesThompson SamplingAndroid
DocMagic
sourcelive demo

Multi-agent RAG assistant that answers questions over logistics, ERP and CRM documents. Every answer is cited to the exact page or sheet.

  • LangGraph supervisor routes each question to one capability specialist (documents, spreadsheet data, arithmetic or live web search), so the model never juggles every tool at once.
  • Hybrid retrieval fuses dense vector search with BM25, then re-scores with a cross-encoder reranker; ingestion runs metadata → analyst → reviewer with a bounded revision loop.
  • Built twice on two branches, Next.js + FastAPI + pgvector and Streamlit + LCEL + Chroma, so the architectural trade-offs are visible side by side.
PythonFastAPILangGraphLangChainNext.jspgvectorLangfuseSSE
Tevel IntelliDB
source

An AI-first SQL client on desktop and Android: a database engineer, not a chat box. Ask in plain English, get safe, reviewable SQL.

  • Built on a fork of Antares SQL (MIT). The Electron and Vue shell is upstream; the AI layer is mine, IntentRouter, SchemaIntelligence, EmbeddingRetriever and SqlValidator, with 36 unit tests. Upstream copyright stays in LICENSE alongside mine.
  • Schema-only RAG: the model reasons over metadata (tables, columns, keys, indexes, relationships) and never sees a single row of customer data. Hybrid retrieval blends a keyword ranker with cached table embeddings by cosine, falling back to keyword-only when embeddings are off.
  • classifyRisk() grades every statement safe, moderate or high, and anything above safe raises a confirm-to-run before it executes. Invalid SQL goes through an auto-repair loop until it validates.
  • A separate Android client, written from scratch rather than forked: Capacitor and Vue 3 over a native JDBC plugin, because a WebView cannot open a TCP socket. MySQL and PostgreSQL through bundled drivers, SQLite through Android's own engine.
ElectronVue 3TypeScriptCapacitorJava (JDBC)MySQLPostgreSQLEmbeddings
ImaGen
source

A prompt-engineering playground for image editing. Pick the technique, see exactly how it changes the result.

  • Eight selectable prompting strategies (zero-shot, few-shot, structured, chain-of-thought, self-refine, meta-prompt, tree-of-thought and raw) compared on the same instruction.
  • Always shows the exact messages sent to the LLM and the final prompt handed to the image model, so the technique's effect is measurable rather than magic.
  • Brush a mask, and the edited frame is composited back over the original inside your region only, working around an image model with no server-side mask.
PythonStreamlitNVIDIA NIMFLUX.2Prompt EngineeringPillow

04 · Skills

What I work with.

LaravelPHPNext.jsReactTypeScriptMySQLMongoDBDockern8nAWSGitHub CI/CDPythonREST APIsMetabaseFastAPILangGraphLangChainRAGMulti-Agent SystemsPGVectorChromaDBLangfuseMCPPrompt EngineeringLaravelPHPNext.jsReactTypeScriptMySQLMongoDBDockern8nAWSGitHub CI/CDPythonREST APIsMetabaseFastAPILangGraphLangChainRAGMulti-Agent SystemsPGVectorChromaDBLangfuseMCPPrompt Engineering

Agentic AI & LLM

Multi-Agent SystemsRAG PipelinesLangChainLangGraphLangFlowMCPPrompt EngineeringHuman-in-the-LoopAgentic Knowledge BasesText-to-SQLAI Document Analysis

AI Engineering Stack

PythonFastAPIPGVectorChromaDBLangfuseHybrid Retrieval (BM25 + Vector)Cross-Encoder RerankingEmbeddingsStreaming (SSE)Streamlit

Backend Development

LaravelPHPREST API DevelopmentCRM & ERP ModulesAPI IntegrationFeature TestingUnit TestingPHPUnitPytest

Frontend Development

Next.jsReactTypeScriptJavaScriptTailwind CSSHTMLCSS

Databases & Search

MySQLPostgreSQLMongoDBSQLManticore SearchFull-Text Search

Async Processing

Laravel JobsEvent-Based ProcessingTask Scheduling

Automation & Workflow

n8nGoogle Apps ScriptZapierAppSheet

DevOps & Deployment

AWSAWS RDSGitHub CI/CDCoolifyDocker

Tools & Platforms

GitGitHubPostmanLinuxMetabase

Languages

PythonPHPJavaScriptTypeScript
Beyond codeProblem SolvingFast LearnerAdaptabilityTeam CollaborationCommunicationLeadershipAnalytical Thinking

05 · Education

Where it started.

Sri Sairam Engineering College, Chennai

B.E. in Electronics & Communication Engineering

2019 - 2023CGPA: 9.05 / 10

06 · Recognition

Research & recognition.

Publications

Patent

An Intelligent Transport Vehicle Management and Routing System

Intellectual Property India

App. No: 202241077102

Filed Dec 2022 · Published Jan 2023

Awards

  • Mentor, Smart India Hackathon (SIH) Finalist Team, 2024
  • Best Outgoing Student, Institution of Engineers India (IEI), 2023
  • School Topper, SSLC (98%), 2017

07 · Contact

Let's build something together.

Whether it's a role, a project, or just an idea worth exploring, I'd love to hear from you. My inbox is always open.

yashwanth.devadoss@gmail.com
GitHub LinkedIn

Speaks English · Tamil