Hi there 👋

I'm Daksh Tyagi.

Computer Science (AI) student in Bengaluru who builds AI-powered software — LLM apps, RAG pipelines and ML classifiers — on reliable, tested backends. Currently Backend Engineer at Decentro.

 
Get in Touch! Grab my resume
AI & Backend Engineer
LLM APIs RAG Agentic AI scikit-learn Python FastAPI Flask SQL Docker AWS React / TS

What I've been shipping on @InnoxCodes.

01.About

I'm a Computer Science engineering student at Manipal Institute of Technology, Bengaluru, specializing in AI. I care about the layer of software people only notice when it fails — backends, data, and the glue between systems.

Right now I'm a Backend Engineer at Decentro, building provider-agnostic payment workflows for a Singapore-based cross-border payments and Banking-as-a-Service platform: multi-currency wallets, payouts, refunds and settlements on Airwallex and DBS rails, backed by a double-entry ledger with idempotency and audit trails.

Before that, at AI Intelligence Systems Lab, I shipped FastAPI services with RAG pipelines and generative voice features. On my own time I build things like TriageAI and ReviewMind — projects where I treat model output as untrusted input and chase honest metrics over flattering ones.

Bengaluru, Indiabased in
Backend Engineer @ DecentroJun 2026 – present
B.Tech CSE (AI), MIT Bengaluru2023 – present · SGPA 8.4
McKinsey Forward Programcertified · Jun 2026
Honeywell & TCS hackathons4th place · Flipkart Grid finalist

02.Experience

Backend Engineer

Decentro · Bengaluru
Jun 2026 – Present

Cross-border payments and Banking-as-a-Service backend for a Singapore-based platform

  • Engineered provider-agnostic payment workflows across multi-currency wallets, payouts, refunds, and merchant settlements, integrating Airwallex and DBS banking rails.
  • Designed and implemented financial ledger architecture — double-entry accounting, balance management, idempotency, and audit trails — supporting risk controls and provider reconciliation.
  • Developed REST APIs for auth, transactions, and merchant onboarding; collaborated with cross-functional teams and clients, and deployed via Docker/AWS (ECR, EC2, CloudWatch).
PythonFlaskMySQLMongoDBSQLAlchemyDockerAWS

AI Engineer

AI Intelligence Systems Lab · Bengaluru
Oct 2025 – Mar 2026

Production AI services, RAG pipelines, and generative voice features

  • Built Python and FastAPI backend services integrating RAG pipelines and external AI components into production workflows, enabling AI-driven features across use cases.
  • Developed and shipped generative AI features using LLM and ElevenLabs APIs, translating product requirements into practical, user-facing solutions.
  • Designed and iterated on AI workflows through rapid development cycles, evaluating outputs and refining system behavior for reliability.
FastAPIRAGLLM APIsElevenLabs

03.Skills

AI / ML

RAG PipelinesLLM OrchestrationAgentic AI WorkflowsPrompt EngineeringAI AutomationScikit-learnPandas

Languages

PythonSQLTypeScriptJSON

Backend

FastAPIFlaskREST APISQLAlchemyWebSocketsSSE

Databases

MySQLMongoDBSQLiteSupabaseDBMSCRUD

Platforms & LLMs

OpenAI APIAnthropic ClaudeHugging FaceElevenLabsVapin8n

Cloud & DevOps

DockerAWS EC2AWS ECRCloudWatchVercelRailway

CS Fundamentals

Operating SystemsComputer NetworksDBMSSystem Integration

Tools

GitGitHubGitHub CopilotPostman

04.Projects

TriageAI

Python · FastAPI · scikit-learn · React/TypeScript
  • Dual TF-IDF + logistic regression classifier for ticket category and urgency, with calibrated confidence, per-token explainability, and low-confidence cases routed to human review.
  • Async FastAPI backend (SQLAlchemy 2.0, SQLite/WAL) with a WebSocket event hub. Caught a template-leakage bug that inflated accuracy to a fake 100% and rebuilt the data generator for an honest evaluation.
  • React/TypeScript Kanban dashboard; both services containerized with Docker and deployed via Railway/Vercel.
94.6%category acc. (vs 22% baseline)
73.2%urgency acc. (vs 31% baseline)
41automated tests

ReviewMind

FastAPI · React/TypeScript · Anthropic/OpenAI APIs
  • Full-stack AI pull-request reviewer that analyzes GitHub diffs file-by-file with an LLM and streams validated findings to a dashboard over Server-Sent Events.
  • Three-layer validation pipeline treating model output as untrusted input: schema enforcement via Anthropic tool use / OpenAI structured outputs, plus domain checks rejecting findings anchored to non-existent lines, with bounded repair retries.
  • Reviewer feedback loop surfacing per-category false-positive rates and per-review token cost.
167tests, fully offline
3-layeroutput validation

AI Voice Receptionist

LLM · Vapi · ElevenLabs
  • AI voice receptionist combining an LLM with speech-to-text and text-to-speech to automate inbound calls and natural-language conversations.
  • Workflows for FAQ handling, appointment scheduling, lead capture, and call routing.
  • Refined prompts and conversation flows across simulated calls to improve intent recognition and response quality.
8+caller intents
50+simulated calls

05.Achievements

Honeywell Hackspace Hackathon

4th PlaceOct 2025

TCS BCIC Hackostav

4th PlaceAug 2025

Flipkart Grid 7.0

FinalistAug 2025

Certifications

McKinsey Forward Program

McKinsey · Jun 2026

Machine Learning for Data Science Projects

IBM · Mar 2026

Intro to Computer Vision & Image Processing

IBM · Oct 2025

Make Foundation

Make Academy · May 2025

Education

B.Tech in Computer Science Engineering (AI)

Manipal Institute of Technology · Bengaluru, Karnataka

Aug 2023 – Present
8.4SGPA

06.Get in Touch

Let's build something that doesn't break.