AstraExplore 3D ◎

M.S. CSE @ UBuffalo · ML/AI Engineer @ Incrivelsoft

Hi, I'm Yasaswi.
I build AI assistants and full-stack products that ship.

ML/AI Engineer @ Incrivelsoft, building Ask NuMaa and community features for numaa.ai. 4+ years across backend, full-stack, and AI/ML work at Incrivelsoft, FactSet, SafetyKnights, and coding.Studio().

Interactive 3DEnter my worldSee projectsAsk AstraRésumé
  • AI Engineer (RAG · LLM Routing · Voice)
  • Full-stack (Angular + React + FastAPI)
  • Test automation (Cypress/Playwright)
Yasaswi Raj Madari

Profile

ML/AI & Software Engineer

AI-ready

Education

M.S. Computer Science (ML/AI Focus), University at Buffalo, completed Jan 2026.

Currently

ML/AI Engineer @ Incrivelsoft, building Ask NuMaa and community features for numaa.ai.

4+
years SWE
5/5
Google Foobar
#17
IEEExtreme 14.0
Astra Lite is waking up

Too lazy to scroll? Let Astra Lite do the talking.

Ask about my numaa.ai work, FactSet systems, test migrations, or production RAG experience. Text or live voice.

Start chatting

OpenRouter-first · Gemini fallback · Voice enabled

Work

Selected work

I thrive at the intersection of product and platform: building experiences that feel sharp, while making sure the system underneath is easy to maintain and extend.

Featured

YashStudio.tech

RAG-powered AI portfolio

Live ↗

Recruiter-facing RAG assistant built with text embeddings, pgvector semantic search, and token-aware context management. Three interaction modes: UI, chat, and voice.

Live at yashstudio.tech with a Netlify frontend and a Render + Neon backend.

  • Built a recruiter-facing assistant that answers questions about my experience using RAG over structured resume and project data stored in Supabase with pgvector embeddings; retrieval is scoped so it does not invent experience that is not in the data.
  • Implemented token-aware context management across multi-turn sessions to keep prompt size bounded instead of re-sending full history; cut average cost per conversation by 40% without losing coherence.
  • Shipped three interaction modes (standard UI, chat, and voice via ElevenLabs TTS/STT) backed by a single Python LLM service that returns schema-validated JSON.
  • RAG
  • FastAPI
  • pgvector
  • OpenRouter
  • Gemini
  • ElevenLabs

02

A.R.M.A

Automated Requests Management Application · coding.Studio()

Case study

Digitized permission requests, letter generation, and hierarchical event approvals with role-based workflows and reusable dashboards.

Cut administrative processing time 40% across 10+ campus forums and 5 departments.

  • React
  • Node.js
  • PostgreSQL

03

Relief Riders

Requests management for Relief Riders Hyderabad

Case study

Dual-portal platform for volunteer intake, triage, and relief-request tracking, with independently deployable requester and admin services.

Built from UML and UI design through implementation; grew to 1,000+ commits, 12 stars, and 16 forks.

  • React
  • Node.js
  • JavaScript
  • System Design

04

Debugging C

Competitive coding platform · IEEE VBIT SB

Case study

Full-stack C programming contest platform with real-time leaderboards. Designed the DB schema, built frontend + backend solo, deployed on PaaS.

Served 500+ participants with 99% uptime and reduced manual judging effort by 80%.

  • Full-stack
  • Database Design
  • C
  • MongoDB
  • Deployment

Skills

What I bring to a team

A blend of computer science rigor and product thinking. I value clarity, observability, and building foundations that scale with the roadmap.

AI / LLM Engineering

RAG, LLM routing, guardrails, persistent user context, voice, and LLM evaluation.

  • RAG Systems
  • LLM Orchestration
  • Agentic Workflows
  • Guardrails
  • LLM Evaluation
  • Context Management
  • PyTorch
  • HuggingFace

Full-Stack Engineering

End-to-end from Angular 19 components to FastAPI services. Frontend polish meets backend reliability.

  • React
  • Next.js
  • Angular 19
  • TypeScript
  • FastAPI
  • NestJS
  • Node.js
  • PostgreSQL

Testing & DevOps

Modernized test suites, cut execution time 35%, and reduced flakiness 20%. Docker, AWS, and CI/CD.

  • Playwright
  • Cypress
  • Docker
  • AWS (EC2/S3)
  • CI/CD
  • gdb
  • UNIX

Experience

How I work with teams

Across AI platforms, financial systems, and full-stack products, I own work end-to-end, collaborate through reviews, and measure the impact of what ships.

  1. ML/AI Engineer

    Incrivelsoft · numaa.ai · California, Remote

    Dec 2025 – Present
    • Built Ask NuMaa with RAG over clinical guidelines, LLM routing, saved chats, and personalized user memory.
    • Shipped Mom to Mom community features and an AI companion, plus streaming voice through ElevenLabs.
    • Deployed independently buildable, testable, and scalable Docker services to AWS with CI/CD.
    More from this role
    • Built Ask NuMaa, the AI pregnancy assistant on numaa.ai: a chat interface with multi-turn conversation history, saved chats, and a user Memory feature, backed by FastAPI and NestJS services with an LLM layer on OpenRouter.
    • Built the RAG pipeline behind it with text embeddings and pgvector semantic search over curated clinical guidelines so answers are grounded in trusted medical content rather than model memory.
    • Implemented question routing with a zero-shot classifier and explicit state machine: off-topic requests are declined, ambiguous ones get a clarifying question, in-scope ones go to the right handler, and users can be handed off to a real doctor when needed.
    • Built the Memory and Add Context features on a MongoDB-backed user context store (profile, pregnancy stage, preferences, prior chats) with embedding-based lookup so the assistant personalizes responses across sessions.
    • Built the Mom to Mom community feature (posts, replies, moderation) and the AI companion chat that runs alongside it, sharing the same auth, user profile, and context store.
    • Integrated ElevenLabs for streaming speech-to-text and text-to-speech so users can talk to the assistant instead of typing.
    • Containerized every service with Docker and deployed to AWS behind a reverse proxy; set up CI/CD so services build, test, and scale independently.

  2. Software Engineer

    SafetyKnights · Buffalo, NY

    Jan 2025 – Jun 2025
    • Owned the Angular 19 frontend and Node/Express backend for an EHS bidding marketplace.
    • Designed indexed PostgreSQL schemas, JWT role-level access control, request validation, and integration tests.
    More from this role
    • Built a bidding marketplace where EHS (environmental, health, and safety) professionals bid on client job postings; owned the feature work across the Angular 19 frontend and Node/Express REST backend.
    • Designed the PostgreSQL schema for users, job postings, and bids, and added indexes for the listing and search queries that showed up as slow in early testing.
    • Implemented JWT-based authentication with role-level access control for client and professional accounts, plus request validation on all write endpoints.
    • Built the listing, bidding, and dashboard views in Angular 19 + TailwindCSS and wrote integration tests for the core API flows before handing the codebase off.

  3. Software Engineer

    FactSet · Hyderabad, India

    Jun 2022 – Jul 2024
    • Developed C++ financial data pipeline modules feeding REST microservices; improved throughput by 25%.
    • Maintained AngularJS UIs and migrated Protractor → Cypress → Playwright, cutting test time 35% and flakiness 20%.
    More from this role
    • Developed and maintained C++ modules in a financial data pipeline that feed REST microservices consumed by client-facing dashboards; optimization work on the data processing paths improved throughput by 25% and cut end-to-end delivery latency.
    • Built and maintained AngularJS 1.x UI components for client-facing financial products, working from JIRA tickets through code review and production release.
    • Debugged production and runtime issues in UNIX environments with gdb and log analysis; managed source through Perforce with JIRA sprint workflows on a 15-engineer team.
    • Migrated the frontend test suite from Protractor to Cypress and then to Playwright, cutting execution time by 35% and reducing flaky tests by 20%, which got CI feedback down to same-day.

  4. Software Development Engineer

    coding.Studio() · Hyderabad, India

    Senior Software Development Engineer · Mar 2021 – May 2022 · SDE III · Jan 2020 – Feb 2021

    Jan 2020 – May 2022
    • Led A.R.M.A and Relief Riders from UML design and UI mockups through full-stack delivery.
    • Owned two major A.R.M.A releases and 1,000+ Relief Riders commits, and mentored junior club engineers.
    More from this role
    • Led full-stack development of A.R.M.A and Relief Riders, two React/Node.js platforms that replaced paper-based request and approval workflows for VBIT and Relief Riders Hyderabad (details under Projects).
    • Owned both codebases end to end, from UML system design and UI mockups through implementation, across two major A.R.M.A releases and 1,000+ commits on Relief Riders.
    • Mentored junior club engineers as Senior SDE: set up code review practices and reusable frontend/backend patterns used by a rotating contributor team.

AI / LLM Engineering

Shipping agentic AI, not just notebooks.

M.S. Computer Science & Engineering (ML/AI Focus) @ UBuffalo. Now building Ask NuMaa and Mom to Mom at Incrivelsoft: clinical RAG, LLM routing and guardrails, MongoDB-backed user memory, community features, and streaming voice.

LLM Orchestration

Built Ask NuMaa with zero-shot question classification, an explicit routing state machine, clarification, guardrails, and doctor handoff.

RAG Pipelines

Built RAG pipelines using text embeddings and pgvector semantic search over curated clinical guidelines and structured portfolio data.

Production AI mindset

Own features from schema and API through frontend, deployment, and tests, with Docker services and CI/CD on AWS.

Mindset

Code + cameras + caffeine.

I ship agentic AI and full-stack features. Survived Protractor, learned gdb, built RAG systems. Also: design + video editing side quests. Buffalo winters built character.

  • End-to-end delivery. Built Ask NuMaa and Mom to Mom, shipped the SafetyKnights marketplace, and reduced FactSet test flakiness by 20%.
  • Full-stack foundation + production AI: retrieval, routing, user memory, community features, and streaming voice.
  • Google Foobar 5/5, IEEExtreme rank 17/4000. Competitive coding meets production software.

Let's talk

Let's build reliable AI and software products.

Currently building numaa.ai at Incrivelsoft. Open to teams that ship fast, care about product, and value reliable AI-enabled systems. I respond quickly.

Email meLinkedInGitHubRésumé