Shashank Karanjkar
Applied AI engineer · Jersey City, NJ

Shashank Karanjkar is a construction engineer building AI for the construction industry.

Previously, Shashank spent nearly five years on construction projects in India and New Jersey, doing the site paperwork by hand. Now he designs and builds AI products around that work, end to end and on his own: retrieval over a firm’s documents, voice-to-data pipelines, and the security and tests that make them safe to hand to a customer.

View selected work Get in touch
01 — Selected work

Two products, one idea: AI takes the office work off the site.

Both are built solo and pre-launch, tested on seeded demo firms. The figures below come from code and test runs, not projections.

A

EasyTab AI

AI back office for small civil contractors

Status
Pre-launch
Role
Founder, solo build
Stack
Next.js, FastAPI, Claude API, Supabase pgvector

An office manager for 2–15 person contractors whose owner is also the estimator, the collections desk and the receptionist.

Each firm gets a private retrieval system over its own contracts, bids and invoices. It answers questions, drafts bid proposals from tender PDFs, writes payment reminders and replies to email. The owner approves every draft before it goes out.

384-dim · local

Retrieval at near-zero token cost. PDF and Word ingestion into ~500-word chunks, local bge-small embeddings, pgvector scoped to each company, and a guard that answers “I don’t have that yet” without calling the model.

9 workflows

One FastAPI service, nine Claude workflows: chat, tender-to-proposal with pricing left blank on purpose, tiered payment reminders, Gmail reply drafting verified live against the Gmail API, and meeting-transcript extraction.

980,473 records

Profiled NYC permit data before building a lead finder. Found owner addresses empty in every row and business names cut at 25 characters, then removed a filter that was rejecting 2,665 opportunities.

Read the case study →
B

FieldLog AI

Voice-first field documentation for QA inspectors

Status
Pre-launch
Role
Founder, solo build
Stack
Next.js, Supabase, Claude, Whisper, offline PWA

Talk for about 90 seconds at the end of the day. The test records, daily report and any non-conformance report are done.

Built from Shashank’s own paperwork as a field engineer: the hour or two a day spent re-typing gauge readings into reports. Speech becomes structured test data, reports fill the firm’s own Word template, and the app works with no signal and syncs later.

2 model tiers

Haiku for structured extraction from voice and photos, Sonnet for reports and chat, behind per-user monthly budgets and a kill switch across 14 AI routes. The model layer can switch to a self-hosted Qwen 2.5 72B and whisper.cpp.

slum → slump

Whisper is primed with construction terms, and a deterministic confusion map flags likely mishearings for the engineer to review instead of changing them silently.

12 tables

Found a cross-tenant hole a green test suite missed because it ran on the admin key. Rebuilt it to sign in as three real accounts, shipped a restrictive row-level policy across 12 tables, and re-verified live.

Read the case study →
02 — Approach

How Shashank builds

i.

Code decides what must be exact

Counts come from the database, prices are left for a person, templates are filled by deterministic code. AI drafts, people approve.

ii.

Test as the real user

Browser tests against a seeded demo firm that assert on model output. Security checks sign in as competing accounts instead of trusting an admin key.

iii.

Spend where it matters

A small model extracts, a stronger model writes, local embeddings keep retrieval nearly free, and every AI call has its own token cap and fallback.

03 — Experience

From job sites to software

  1. 2026 – Present
    Founder and solo builderEasyTab AI and FieldLog AI
  2. Sep 2025 – Present
    Project EngineerJZN Construction, NJ
  3. Jul 2025 – Sep 2025
    Project ManagerSalvi Steel, NJ
  4. May 2025 – Jul 2025
    Project ManagerNortheast Stage, LLC, NJ
  5. Aug 2024 – May 2025
    Operations AssociateColumbia University, NY
  6. Jul 2019 – Dec 2022
    Assistant Manager, ConstructionS R K Food & Fuel Pvt Ltd., India
Education
M.Eng, Engineering Management — Stevens Institute of Technology, 2024
B.E., Civil Engineering — Datta Meghe College of Engineering, Mumbai, 2021
Toolkit
AI and LLM

Claude API, RAG, embeddings and pgvector, Whisper, Claude vision, LLM output testing with Playwright, Claude Code

Software

Python, TypeScript, SQL, Next.js, React, FastAPI, PostgreSQL, Supabase with row-level security, offline PWAs

Integrations

Gmail API with OAuth 2.0, Google Docs, Drive and Sheets, WhatsApp Cloud API, Telnyx, SAM.gov, NYC Open Data

Construction

AutoCAD, Revit, Primavera P6, Microsoft Project, Trimble Unity Construct, Lean and Six Sigma

04 — Contact

Talk to Shashank about AI that works on a job site.

Shashank is open to applied AI engineering roles in the New York area, and glad to talk with contractors and construction teams about the products.

LinkedIn
linkedin.com/in/shashank-karanjkar