Skip to main content
Rohit Raj
HomeAgentsProjectsServicesReposNotesAboutContactView Current Work

Founding engineer for hire in India — ships end-to-end on any stack (AI, backend, mobile, blockchain, iOS, web3) without the cost of a team.

Founding Engineer for Hire in India.
AI MVP live in 6 weeks — 29 products shipped, every one documented.

Book free 30-min callAI ProjectsEngineering Notes
  • Senior engineer · GitHub from day one
  • You own the code · MIT or commercial
  • Daily Slack / WhatsApp access
  • First production commit in 5 days
AI and Backend Systems Architecture
Founding Engineer for Hire

A founding engineer for hire in India — without the cost of a team.

Most early-stage founders need one senior engineer who can ship the whole stack: AI, backend, mobile, infra, and the awkward integrations in between. That is exactly the gap a founding engineer for hire in India closes — at 30–40% of the cost of a US contractor and with daily timezone overlap on Slack or WhatsApp.

  • →Senior full-stack + AI engineer, based in Pune, India — shipping to startups worldwide.
  • →End-to-end ownership: backend, frontend, mobile, infra, observability, deployment.
  • →Production AI MVP live in 6 weeks; first commit pushed to your GitHub on day 5.
  • →You own the code, the IP, and the repo from week 1 — MIT or commercial license.
Book a free 30-min call
AI Projects

Production-Ready AI Systems

Not experiments — full-stack applications with real infrastructure.

01

MyFinancial — Personal Financial Advisor

liveLiveSource
MyFinancial — Personal Financial Advisor

Problem

Financial planning in India is fragmented across banks, insurance, and tax documents. Most tools require sharing sensitive data with third parties.

Solution

Privacy-first PWA that consolidates financial data locally via a 6-step wizard — Profile, Income, Assets, Liabilities, Insurance, Tax — with real-time advisory metrics like Financial Runway and Savings Rate.

AI Approach

Rule-based advisory engine for Indian financial instruments (PPF, EPF, NPS). Old vs. New Tax regime comparison. Coverage gap analysis for insurance. No cloud dependency — all computation runs locally.

Data privacy
100% on-device
Wizard completion
6 steps · ~4 min
Tax regimes covered
Old + New
React 19Vite 7Tailwind CSS 4ZustandDexie (IndexedDB)Spring Boot 3.xJava 21PostgreSQL
02

PropCheck — AI Property Trust Score for India

liveLiveSource
PropCheck — AI Property Trust Score for India

Problem

Indian property buyers lose lakhs to fraudulent listings on Magicbricks, 99acres, Housing.com, and NoBroker. Fake RERA numbers, recycled stock photos, and inflated pricing slip past buyers because no neutral tool exists to verify a listing in seconds.

Solution

Paste any listing URL — the AI engine scrapes the page (with an LLM parsing fallback when sites are SPA or rate-limited), cross-checks 8 trust signals against Karnataka RERA, a locality price index, and a perceptual-image database, and returns a 0–100 Trust Score with explainable red flags in 30 seconds.

AI Approach

8-signal trust engine — listing age, price-vs-locality delta, duplicate count, RERA registration check, image reverse-search via perceptual hashing, builder complaints, owner-name match, suspicious patterns. Gemma 4 31B via OpenRouter free tier kicks in as LLM parsing fallback when scrapers fail.

Trust score
0–100 in 30s
Signals checked
8 per listing
API endpoint
api.rohitraj.tech
Next.js 14Tailwind CSSFastAPI 0.115Python 3.12PostgreSQL 16SQLAlchemy 2httpxBeautifulSoup4imagehashOpenRouter (Gemma 4 31B)Chrome MV3
03

StellarMIND — Chat-to-SQL with pgvector

liveSource
StellarMIND — Chat-to-SQL with pgvector

Problem

Business users need to query databases without knowing SQL. Existing tools lack context-aware query generation and safety guarantees.

Solution

Spring Boot MCP server that converts natural language questions into read-only SQL using LLM with retrieval-augmented context from pgvector.

AI Approach

RAG-based SQL generation: schema knowledge stored as embeddings in pgvector, retrieved as context for LLM. Strict read-only enforcement (only SELECT/WITH).

Query latency p95
<1.2s
SQL safety
100% read-only
Schema embeddings
pgvector
Spring BootSpring AIPostgreSQLpgvectorMCP ProtocolOpenAI
04

MicroItinerary — AI Travel Planner

liveSource

Problem

Travel apps optimize for proximity and ratings. They don't consider human energy levels, group dynamics, or budget constraints intelligently.

Solution

AI-powered PWA that generates personalized annual travel itineraries with intelligent destination suggestions, cost estimation in INR, and Splitwise-style expense splitting.

AI Approach

GPT-4 for destination recommendations based on season, budget, and preferences. AI-generated cost breakdowns for hotels, food, transport, and activities.

Build time
6 weeks
GPT-4 cost / itinerary
<$0.08
PWA Lighthouse
94/100
React 18ViteSpring Boot 3.2.2Java 21PostgreSQL 16RedisOpenAI GPT-4
Read full architecture notes → →
AI Agent Host

Autonomous agents, aimed at billion-dollar markets

Not chatbots — agents that decide, call tools, and finish a job on their own. Seven of them run live in the Agent Host.

01

Resolvr — Autonomous Support Resolution Agent

liveTry it live SourceSelf-hosted AI support agent — full guide →
Billion-dollar marketAutonomous customer-support resolution$15.1B (2026) → $117.9B (2034), 25.8% CAGR · Gartner: 80% of routine support AI-handled in 2026

Problem

Support teams drown in repetitive tier-1 tickets — password resets, invoices, how-tos. Hiring to keep up is expensive, queues blow out, and naive auto-replies hallucinate policy or promise refunds no one approved.

What the agent does

A standalone full-stack product (FastAPI + React) that takes a raw support ticket to a finished outcome on its own: it classifies the request, retrieves the right knowledge-base articles by semantic search, decides whether it can safely resolve or must escalate, and drafts the reply grounded strictly in the KB — never inventing policy and never promising a refund.

Autonomy

A four-tool loop — classify_ticket → search_kb → decide_action → draft_resolution. Retrieval is real RAG (Ollama embeddings + cosine); the reply is written by a local Ollama model (qwen2.5:14b), falling back to a cloud API key, then to a template, so it self-hosts at zero per-token cost. A hard safety gate forces escalation on security, legal, abuse, and refund tickets — proven at 100% must-escalate recall by the pytest eval suite. Try the lite in-browser demo below; the full product runs separately.

Resolvr — Autonomous Support Resolution Agent — running screenshot
Running standalone — React UI, FastAPI backend, reply drafted live by local Ollama.
Must-escalate recall
100%
LLM backend
Ollama + API
Stack
FastAPI + React
FastAPI (Python)React + ViteOllama + cloud-API fallbackSemantic RAG (embeddings)SQLitepytest eval-gated
02

Loopr — Self-Improving Prompt-Optimization Loop

liveTry it live Source
Billion-dollar marketLLM evaluation & prompt optimization (LLMOps)Fastest-growing dev-tool category of 2026 · GEPA-style reflective evolution beats RL by 20% with 35x fewer rollouts

Problem

Every LLM team tunes prompts by hand — change a word, eyeball a few outputs, repeat. There is no cheap, local way to answer "which prompt actually works best for this task?", and the eval incumbents (MLflow, Confident AI, Opik, LangWatch) are heavyweight hosted platforms you have to adopt.

What the agent does

A drop-in loop that takes a task prompt plus a handful of eval cases and evolves the prompt on its own: it runs the prompt, scores every output, reflects on the failures in plain language, rewrites the prompt, and repeats until it converges on the best-performing version — emitting a transparent trace of every iteration.

Autonomy

A run → score → reflect → rewrite loop (GEPA-lite reflective evolution). Scoring and the stop decision are deterministic and eval-gated — the LLM only runs the task and phrases the reflection — so a run is reproducible and a bad rewrite can never corrupt state. It stops on converged, plateau, or budget. Five scorers (exact, contains, regex, json_field, llm_judge); local-first on Ollama (qwen2.5) with OpenAI/Anthropic fallback at zero per-token cost. Proven 0% → 100% on a JSON-extraction task in one iteration; a 28-case pytest suite gates the deterministic core.

Prompt lift (demo)
0% → 100%
Tests passing
28 / 28
Cost / run
$0 local
PythonOllama + cloud-API fallbackGEPA-lite reflective loopDeterministic scorerspytest eval-gatedstdlib-only HTTP
03

Dispatchr — Autonomous Home-Services Dispatcher

liveTry it live Source
Billion-dollar marketAI front desk for home services$600B+ US home services · AI receptionists a breakout 2026 category

Problem

Home-services businesses (HVAC, plumbing, electrical) bleed revenue from missed calls and slow replies. A round-the-clock human dispatcher is expensive, and after-hours leads simply go cold.

What the agent does

An agent that takes a customer from "I have a problem" to a booked appointment on its own — it classifies the job, quotes only from a real price book, offers genuine open slots, and books the technician. No human in the loop for the happy path.

Autonomy

A transparent tool-calling loop: classify → get_price_estimate → find_available_slots → book_job. A hard safety rule overrides everything — any hint of gas, fire, smoke, sparks, shock, or flooding triggers an immediate hand-off to a human. Decisions are deterministic and gated by a 26-case eval suite.

Eval pass rate
26 / 26
Emergency recall
100%
Over-escalation
0%
TypeScriptNext.js API routeTool-calling loopEval gateOpenAI-compatible (swappable)
04

ClauseGuard — AI Contract Risk Review

liveTry it live
Billion-dollar marketContract review for freelancers & SMBs$10B+ legaltech · every freelancer and small business signs contracts they never fully read

Problem

Freelancers and small businesses sign NDAs, MSAs, and SaaS terms they never fully read — then get caught by uncapped liability, IP over-assignment, non-competes, auto-renewals, and Net-90 payment traps.

What the agent does

A first-pass reviewer that reads a contract, flags each risky clause with the exact offending quote, explains in plain English why it matters, and proposes a concrete redline — and it also flags protective clauses that are missing entirely.

Autonomy

A deterministic 16-rule playbook plus absence checks runs offline with no API key, ranks every finding high / medium / low, and produces a stable risk grade. The same playbook anchors the eval suite.

Clause rules
16 +
Categories
9
Runs
offline
TypeScriptNext.js API routeRule playbookOffline-firstEval-gated
05

FinScope — Portfolio X-Ray (educational)

liveTry it live Source
Billion-dollar marketPersonal-finance diagnostics for IndiaCrores of Indian mutual-fund investors · almost none audit overlap, cost, or tax drag

Problem

Retail investors hold overlapping funds, pay above-median expense ratios, trip short-term capital-gains taxes, and run concentrated or under-cushioned portfolios — with no easy way to see any of it.

What the agent does

An X-ray that scores a portfolio across six dimensions — allocation drift, fund overlap, expense drag, tax efficiency, concentration, and emergency fund — and turns every issue into a specific question to take to a SEBI-registered advisor.

Autonomy

Deterministic analysers crunch the numbers while a hard compliance gate guarantees the output never says buy, sell, or switch. It flags and explains — every flag ends with a question for a SEBI-registered RIA, never a recommendation.

Health checks
6 dims
Compliance
0 violations
Advice given
never
TypeScriptNext.js API routeDeterministic analysersCompliance gateEval-gated
06

MCPGuard — MCP Manifest Security Scanner

liveTry it live Source
Billion-dollar marketSecurity for the agentic / MCP supply chainAI-agent security a breakout 2026 category · every third-party MCP tool is new attack surface

Problem

AI agents now load third-party MCP tools whose manifests can carry hidden prompt injections, exfiltration directives, embedded secrets, or shell access — and almost nobody scans them before wiring them into an agent.

What the agent does

A scanner that statically inspects an MCP manifest and flags the dangerous patterns — prompt injection in descriptions, tool poisoning / cross-tool hijacks, over-permissioned shell commands, leaked credentials, wildcard scopes, and unauthenticated dangerous tools — each with a concrete fix.

Autonomy

Six deterministic check families run with zero network and no LLM, grade the manifest A–F, and return the exact evidence that tripped each rule. A faithful port of a Python core that scores 100% recall on its eval suite.

Threat classes
6
Eval recall
100%
Runs
no API key
TypeScriptNext.js API routeStatic analysisZero-networkEval-gated
07

Cadence — Autonomous SEO Content Agent

liveTry it live Source
Billion-dollar marketProgrammatic content marketing$400B+ content marketing · AI is rewriting how brands win organic traffic

Problem

Content marketing stalls on the boring middle: drafting publish-ready posts, hitting SEO structure, and keeping quality consistent at volume. Hand-written posts don't scale; AI drafts are inconsistent and often skip the on-page SEO that actually ranks.

What the agent does

An agent that takes a topic to a publish-ready SEO post on its own — title, meta, slug, body, FAQ, and JSON-LD schema — then runs a structural linter that grades it pass/fail before it ever ships. The same topic always produces the same audited draft.

Autonomy

A four-tool pipeline: pick_topic → draft_post → validate_seo → save_post. The linter runs a 10-point structural check and the agent auto-revises once on failure. Fully deterministic, no API key — gated by a quality suite covering SEO validity, keyword placement, and schema.

SEO validity
100%
Schema validity
100%
Runs
no API key
TypeScriptNext.js API routeContent pipelineStructural linterEval-gated
08

Prospectr — Autonomous Outbound BD Agent

liveTry it live Source
Billion-dollar marketOutbound sales & lead generation$30B+ sales engagement · outbound is mostly manual, spammy, and low-conversion

Problem

Outbound BD is a grind: verifying emails, scoring whether a lead even fits, and writing a pitch that doesn't read like a template. Done by hand it's slow; done by naive automation it's spam that torches sender reputation.

What the agent does

An agent that takes a raw lead to a personalized, queue-ready pitch — it verifies the email, scores fit 0–100 against a fixed ICP, and only for keepers drafts a ≤140-word pitch with no placeholder leaks. A blocklist gate suppresses bad domains before anything is queued.

Autonomy

A four-tool pipeline: enrich_lead → score_fit → draft_pitch → queue_send. Sending is dry-run by design — it physically cannot transmit — and a safety gate suppresses blocklisted domains. Deterministic and eval-gated on fit accuracy, personalization, and blocklist suppression.

Fit accuracy
100%
Blocklist suppression
100%
Placeholder leaks
0
TypeScriptNext.js API routeICP scorerSafety gateDry-run only
09

Founder-Agent — Autonomous Startup Operator

active
Billion-dollar marketTurning a fintech audience into a productIndia personal finance · target 10M weekly users, ₹1,000 Cr+ ARR

Problem

Solo founders stall in the gap between insight and execution. Strategy work is endless, easy to procrastinate, and rarely compounds into something the next day can build on.

What the agent does

An autonomous operator that, every run, reads the mission, picks the single highest-leverage next move, and ships a concrete artifact — a spec, funnel copy, a validation experiment. The artifacts stack up; each one is something a competent operator could execute tomorrow.

Autonomy

A DECIDE → EXECUTE → COMMIT loop with forced-JSON decisions, an append-only journal, and persistent state memory. Runs 100% locally on Ollama — no API key, no cloud, no per-token cost — so it can grind indefinitely.

Runs
100% local
Artifacts shipped
5+
Cost / run
$0
PythonOllamaqwen2.5 / hermes3Forced-JSON tool useLocal-first
10

GEO Engine — Generative Engine Optimization

development Source
Billion-dollar marketGetting brands cited by AI search$1.48B → $17B by 2030 · 45.5% CAGR

Problem

Classic SEO is collapsing — 93% of AI answers are zero-click and rarely cite the brands behind them. Companies are going invisible inside ChatGPT, Perplexity, and AI Overviews. The funded incumbents only measure that invisibility; they don't fix it.

What the agent does

An execution-layer agent that auto-generates and publishes AI-citation-bait content — comparison pages, structured Q&A, schema markup — then measures citation lift. It closes the gap incumbents leave open: actually making a brand answerable, not just dashboarding the damage.

Autonomy

Built on the same zero-CAC content engine that already ranks my own sites organically — the pipeline researches a topic, drafts citation-optimized content, publishes, and tracks whether AI engines start citing it.

Market CAGR
45.5%
TAM by 2030
$17B
Crosses $1B
2027
Next.jsTypeScriptContent pipelineStructured data / schemaCitation tracking
Open the Agent Host — try a live agent →
Process

How a 6-week MVP sprint works

Fixed scope. Daily Slack. First production commit by day 5.

  1. 01
    Week 1

    Discovery & architecture

    • Problem framing call (90 min)
    • Architecture doc + tech stack lock-in
    • Milestone schedule signed
  2. 02
    Week 2

    Core backend & auth

    • Database schema + migrations
    • Auth flow (email / OAuth)
    • First production deploy on day 5
  3. 03
    Week 3

    AI / data layer

    • LLM integration (RAG / agents)
    • Vector store + retrieval pipeline
    • Cost guardrails + token budgeting
  4. 04
    Week 4

    Frontend & UX

    • UI flows + mobile responsive
    • Onboarding + empty states
    • Analytics events wired
  5. 05
    Week 5

    Hardening

    • Load tests + p95 latency budget
    • Observability + alerting (Prometheus + Grafana)
    • Security pass + dependency audit
  6. 06
    Week 6

    Launch

    • Bug bash + UAT
    • Public deploy + DNS cutover
    • Code handover + runbook + docs
Engineering Quality

Reliability & Production Readiness

📊

Observability

Prometheus + Grafana

Production-grade metrics, dashboards, and SLO visibility.

  • RED/USE metrics with custom business KPIs
  • Grafana dashboards for latency, throughput, error rates
  • Alerting and environment-aware scrape configuration
Learn more →
⚡

Load Testing

k6

Performance validation for high-throughput, event-driven systems.

  • Scenario-based tests (ramping, soak, constant-arrival-rate)
  • Thresholds on p95/p99 latency and error rates
  • CI-compatible execution and reports
Learn more →
🔗

API Contract Testing

Postman + Newman

Repeatable regression and smoke testing for REST APIs.

  • Environment-driven collection execution
  • Newman CLI with HTML/JUnit reports
  • Pipeline-friendly contract validation
Learn more →
📨

Event-Driven Testing

Kafka Simulation

Deterministic testing of Kafka consumers and workflows.

  • Forked Kafka simulation repos for event replay
  • Partitioning and ordering validation
  • Failure, retry, and backpressure testing
Learn more →
Testimonials

What Clients Say

Rohit delivered our MVP in 5 weeks — on budget and ahead of schedule. His architecture decisions saved us from rewriting everything when we scaled.

Arjun Kapoor
Founder, NovaByte Labs
MVP Development

We needed a WhatsApp bot for our clinic chain. Rohit understood the problem immediately and shipped a working solution that our staff could use without training.

Priya Mehta
CTO, MediConnect Health
WhatsApp Bot

What impressed me most was the transparency. GitHub access from day one, weekly demos, no surprises. The React Native app he built is still running with zero issues.

Vikram Desai
Product Manager, FinLeap Technologies
Mobile App
FAQ

Common questions before we start

Why hire a founding engineer for hire in India over a US contractor?

A senior founding engineer for hire in India ships the same production AI MVP at 30–40% of US contractor cost, with full code ownership and daily timezone overlap on Slack/WhatsApp. You get end-to-end execution — backend, frontend, mobile, infra, AI — from one person who has shipped 29 products.

How fast can a founding engineer in India ship my AI MVP?

First production commit on day 5, fully-live MVP at week 6 — fixed scope, fixed timeline, no "agency surprise". Every project on rohitraj.tech follows this 6-week shipping cadence.

What if 6 weeks slips?

Fixed scope means we descope features, not extend timeline. You ship on week 6, even if a few stretch features move to v2. You keep what's working.

Who owns the code?

You. Full repo handover on week 6. MIT license by default; commercial-only if you prefer. All IP transfers on final payment.

Am I locked into your tech stack?

No. I default to Next.js / Spring Boot / Postgres because they're boring and hireable. If your team uses Python / Go / Rails, we ship in your stack.

What about post-launch maintenance?

Two options: (1) clean handoff to your team with full docs and a runbook, or (2) ongoing retainer at a fixed monthly rate. No surprise bills either way.

Are you the only engineer?

Yes. No outsourcing, no rotating contractors. You talk to me on Slack / WhatsApp daily. If I get overcommitted I tell you upfront.

Are you open to long-term collaboration?

Yes. Most work starts as a focused sprint, then can continue as a retainer, advisory role, or deeper technical partnership if the product and team are a strong fit.

NDAs and IP?

Standard mutual NDA before kickoff if you want one. All code IP transfers to you on final payment, signed in the master services agreement.

Rohit Raj — Backend & AI Systems

Services

AI Agent HostFounding Engineer for Hire in IndiaMobile App DevelopmentAI Chatbot DevelopmentFull-Stack Development

Get Updates