(PS—26) sample · seed 2026 denoising · step 00 / 50 currently: AI-chat · reAlpha
Hi, I’m

Prajesh Shrestha

I build chatbots and agents that people actually use, and I’m still learning how.

See the work ↓ Say hello
waiting for the sample to settle…See all the work ↓
2024 · reAlphaProperty tags— 2025 · reAlphaClaire— 2025 · still liveOnboarding Agent 1.0— 2025–26 · reAlphaMortgage agents— 2026 · ~1M users/moAI-chat— 2023–24 · B.E. projectCrossword solver—
(02) So far

Learning curve

time not to scale

It goes up, mostly. Every new chapter starts with a dip, and that's usually where the learning is.

what I know (roughly) ↑
  1. 2019Started B.E. Computer EngineeringEverything was new. Started from roughly zero.
  2. 2023Crossword solver, Heritage LensFirst time a model I trained did something I couldn’t.
  3. Aug 2024Joined reAlpha · property tagsProduction data is nothing like a dataset.
  4. 2025Claire, Onboarding Agent 1.0Shipped an agent people actually use. Still live.
  5. Oct 2025Mortgage agentsLong workflows: design for the step where people get stuck.
  6. May 2026AI-chatNow I look after the AI layer with two teammates. A new dip, a steeper climb.
(03) Evals

A few numbers, with context

mostly team results
Monthly users, AI-chat
~1M
Across client deployments of the platform.
Responsible for the AI layer, team of three
Onboarding Agent 1.0
Live
Still onboarding new users on reAlpha since 2025.
Built and delivered it
Team
3
Me, a junior and a mid-level engineer.
I lead delivery and review
＋ Add an eval you’re proud of (e.g. resolution rate before/after, latency p95)
(04) Career

Now, and how I got here.

CHAPTER III · NOW May 2026 — Present

Senior AI/ML Engineer

reAlpha · AI-chat

I lead the AI behind AI-chat, a Singapore-based B2B platform of customized chatbots we build for client businesses.

  • Own the AI system end to end and take responsibility for how it behaves in production
  • Lead a junior and a mid-level engineer from scoping through delivery
  • Enhance the existing platform, used by 1M+ people every month
＋ Add a headline result (e.g. latency, resolution rate, cost)
LeadershipLLM systemsB2B chatbots
CHAPTER II Oct 2025 — May 2026

AI/ML Engineer, Level II

reAlpha · Mortgage agents

Built mortgage agents: LLM-driven assistants that guide people through the lending process one step at a time.

  • Designed conversational workflows across the mortgage journey
  • Personalized guidance at each step of an application
＋ Add impact (e.g. applications assisted, drop-off reduced)
AgentsTool useWorkflows
CHAPTER I Aug 2024 — Aug 2025

AI/ML Engineer, Level I

reAlpha · Naamche · Property tags · Claire · Onboarding Agent 1.0

My first year in production AI, spread across three products on the main reAlpha platform.

  • Built property tag generation and tagging for listings in the realty service
  • Enhanced Claire, the AI chatbot the team built for the main site
  • Delivered Onboarding Agent 1.0, optimized for onboarding users and still live today
＋ Add a metric for Onboarding Agent 1.0
RAGChatbotsNLPTagging
PROLOGUE 2019 — 2024

B.E. Computer Engineering

Thapathali Campus, Tribhuvan University · Deep learning & NLP

Where I fell for language and vision. My major project taught a machine to solve and generate crosswords.

  • Specialized in deep learning and NLP
  • Major project: Crossword Solver & Generator
  • Heritage Lens: monument detection with computer vision
＋ Add GPA / honours / thesis link
Deep learningNLPVision
(05) Academic & research

Research

Where it started: language, vision, and a stubborn interest in puzzles machines can’t solve yet.

(R1) · B.E. major project · Jun 2023 — May 2024

Crossword Solver & Generator

NLPConstraint satisfactionDeep learning
PROBLEM
＋ What made crosswords hard for machines?
APPROACH
An intelligent system using NLP, constraint satisfaction, and deep learning to both solve and generate crossword puzzles.
OUTCOME
＋ Accuracy / grids solved / demo link
Crossword Solver & Generator
(R2) · Academic research · Jan — May 2023

Heritage Lens

Computer visionDeep learning
PROBLEM
＋ Why detect monuments automatically?
APPROACH
A computer-vision and deep-learning pipeline that automatically detects monuments within images.
OUTCOME
＋ mAP / dataset size / demo link
Heritage Lens
(R3)＋ Publication / paper · venue · year
(06) Personal

Builds

Things I made to understand something properly. Hover a row.

(B1)SortosA visualization tool for understanding and comparing sorting algorithms in action.VIZ (B2)A* Pathfinding VisualizerInteractive visualization of A* pathfinding, demonstrating optimal path discovery.ALGO
(B3)＋ Next personal project · one-line description · link
(07) Recognition

Awards & certifications

(A1)＋ Title · issuer · year
(A2)＋ Title · issuer · year
(A3)＋ Title · issuer · year
(08) Contact Back to the latent space.

Say hello.

Building something with LLMs, or need someone to own the AI part? I’d love to hear about it.

＋ your@email GitHub LinkedIn CV