000%

Great things take time

Please wait
Available for internships & freelance

>

I build AI systems that negotiate, remember, and ship — from self-hosted LLM runtimes to blockchain escrow.

Mangaluru, Karnataka, India
About

Engineer who ships systems, not just screens

I'm a Computer Science & Business Systems undergrad who treats side projects like production. I gravitate to the hard middle layer — inference infra, smart contracts, real-time pipelines — where good engineering actually shows.

The short story

I started where most do — building UIs that looked good. The obsession crept in when I realized the interesting problems live underneath: how do three AI agents reach a fair deal? How do you serve an LLM without a cloud bill that ends your weekend? How do you guarantee a ledger never double-counts?

Now I build end-to-end. Accord runs competing agents that settle escrow on Algorand. EternoMind compresses conversation history into memory banks to route requests to cheaper models. FeedForward scores plate waste with computer vision in milliseconds. Different domains, same instinct — make the system smarter so the user does less.

When I'm not coding I'm leading the Automation Robotics Club, automating campus labs, or winning the next hackathon.

Education

  • St Joseph Engineering College, Mangaluru

    B.E — Computer Science & Business Systems

    Aug 2023 – PresentCGPA: 9.3 / 10.0

  • St Mary's PU College, Kundapur

    Class XII · KSEAB — Science (PCM + CS)

    March 202389.6%

  • Holy Rosary English Medium School, Kundapur

    Class X · KSEAB

    July 202192.8%

AI Systems

Self-hosted LLM runtimes, agent negotiation, and memory-aware inference routing that actually cuts cost.

Full-Stack Velocity

React, Next.js and FastAPI from prototype to production — fast, typed, and shipped to real users.

Systems Thinking

From 7-state order lifecycles to on-chain escrow, I design for correctness, not just the happy path.

Tech Stack

Tools I reach for

A categorized look at the languages, frameworks and infrastructure behind my projects. Hover a category to explore.

Languages

8 skills
TypeScript90%
Python92%
Java85%
JavaScript90%
Kotlin75%
Dart78%
C / C++80%
Bash72%
TypeScriptPythonJavaJavaScriptKotlinDartC / C++BashReact.jsNext.jsTailwind CSSViteNode.jsExpress.jsFastAPILangGraphvLLM / ROCmYOLOv8FlutterAndroid StudioPlatformIO / ESPPostgreSQLMongoDBRedisFirebaseMySQLSupabase / NeonDockerVercelGit / GitHubCloudinaryLinuxTypeScriptPythonJavaJavaScriptKotlinDartC / C++BashReact.jsNext.jsTailwind CSSViteNode.jsExpress.jsFastAPILangGraphvLLM / ROCmYOLOv8FlutterAndroid StudioPlatformIO / ESPPostgreSQLMongoDBRedisFirebaseMySQLSupabase / NeonDockerVercelGit / GitHubCloudinaryLinux
Featured Work

Projects built end-to-end

Three systems where the hard part wasn't the UI — it was the infra, the agents, and the guarantees underneath.

accord.tsx
Accord
01AI-Negotiated Blockchain Escrow Platform

Accord

An autonomous escrow platform where three competing AI agents negotiate deals through a weighted optimizer, while an ARC-4 smart contract handles fund locking and release on Algorand — no human intermediary required.

6 in ~4s
Negotiation turns
7× faster
vs. 30s target
3
Competing agents
  • Three competing agents negotiate deals via a weighted optimizer with autonomous on-chain settlement.
  • Self-hosted LLM inference stack (vLLM + ROCm) with a 3-tier fallback chain for reliability.
  • ARC-4 smart contract handles trustless fund locking and release on Algorand.
ReactNode.jsPythonAlgorandFastAPIDockervLLM
eternomind.tsx
EternoMind
02Self-Optimizing Memory-Aware AI Runtime

EternoMind

A full-stack AI runtime that cuts LLM inference cost by compressing user-specific conversational context into persistent memory banks, then dynamically routing requests to cheaper models once memory coverage is sufficient.

up to 60%
Inference cost cut
11
Orchestration nodes
SSE realtime
Streaming
  • Compresses conversational context into persistent Hindsight memory banks to reduce cost.
  • 11-node LangGraph pipeline: memory retrieval, RAG, prompt optimization, routing, validation, writeback.
  • FastAPI SSE streaming backend with a React dashboard for token, latency and savings metrics.
PythonFastAPILangGraphReactGroq
feedforward.tsx
FeedForward
03Food Waste Reduction Platform

FeedForward

A food-waste tracking platform using YOLOv8 with a classical CV fallback to analyze plate waste in real time, paired with a penalty ledger that routes the majority of collected funds to NGO partners.

<200ms
Scan latency
95%
Funds to NGOs
7 states
Order lifecycle
  • YOLOv8 + CV fallback (Hough circles + contour analysis) analyzes plate waste in under 200ms.
  • Penalty ledger charges restaurants and routes 95% of collected funds to NGO partners.
  • Zero double-counting enforced across a 7-state order lifecycle.
ReactNode.jsExpressPythonFastAPIYOLOv8
Experience

Where I've shipped

Leadership, internships and the automation work that keeps a campus running.

Part-time · Leadership

Software Lead

Automation Robotics Club (ARC), SJEC

Sep 2024 – PresentMangalore, India
  • Built an RFID-based door access and attendance system with real-time Firebase sync, reducing manual attendance logging to zero across 2 campus labs.
  • Delivered an automated bell scheduling system (ESP01S + relay) and a QR-based inventory tracker, cutting manual bell operation and stock reconciliation time for the department.
Internship

Generative AI Virtual Intern

Skillible

Jul 2024 – Aug 2024Remote
  • Completed 15+ structured prompt engineering tasks covering few-shot prompting, chain-of-thought reasoning, and output constraint techniques across text and code models.
  • Analysed AI model behaviour across 5 bias categories (gender, cultural, recency, confirmation, framing), documenting failure modes and mitigation strategies in a capstone assessment.
Achievements

Hackathons & competitions won

Six wins and counting across national and intercollegiate events — embedded systems, ideathons and full-stack builds.

1st

Aerithon26 Hackathon

Winner — 1st place out of 70+ teams. Canara College of Engineering, Mangaluru.

2026

2nd

Xypheria 2026

2nd place at a national-level technical competition. Alva's Institute of Engineering & Technology.

2026

1st

Magnovite 2025 — Christ University

1st in Bytes & Boards (Embedded Systems) and 1st in Enigma Redesign.

2025

1st

Magnovite 2024 — Christ University

1st in The Gentleman Burglar and 2nd in Enigma Redesign.

2024

4×

Intercollegiate Ideathons

Won across 4 competitions (Canara, SJEC, SIT, Sri Devi) out of 200+ teams overall.

2024–25

Open Source

Building in public

Most of my work lives on GitHub — from LLM runtimes to embedded firmware. Here's the activity at a glance.

++++

Activity_Heatmap

@Preethesh16
++++
—

DAYS ACTIVE

++++
—

LONGEST STREAK

CONSECUTIVE

++++
—

CURRENT STREAK

TRAILING

++++
—

TOTAL COMMITS

OCT 2024–2026

MON
WED
FRI
LESSMORE
Contact

Let's build something

Open to internships, freelance, and collaborations. Drop a message — I usually reply within a day.

Get in touch

Whether it's a role, a project, or just to talk LLM infra and hackathons — my inbox is open.

preetheshcarvalho57@gmail.com
Mangaluru, Karnataka, India

Find me online