DEEPAK·SAHARAWAT
Available — contract & fractional

AI Systems Builder · Bangalore, India

I build and ship production AI systems, end⁠-⁠to⁠-⁠end.

Eighteen years as an operator and founder; the last two building production AI SaaS solo — retrieval, LLM orchestration, search, data, billing, and deployment. I don't stop at prototypes — I ship systems that run under load, and I can explain every decision in them.

18+
years operating & building
6
production-grade AI products
1 live
flagship in production
750K+
document-chunk RAG corpus
360+
test files in the live flagship

Flagship — live in production

Vidhi

A domain-specific AI research & drafting platform — RAG over a very large, citation-heavy corpus. Soft-launched with real users and live subscription billing. Sole architect and engineer.

Production RAG platform

Co-Founder & CTO · Feb 2024 – present · sole architect & engineer
Live / soft-launched

Fast, source-cited answers over 750,000+ document chunks — with the full product built around the model, not just an API wrapper.

Retrieval
Hybrid search fusing lexical (BM25) and vector (pgvector) results via reciprocal-rank fusion. Source-cited answers, follow-up context, stable quality as the corpus grows.
LLM orchestration
Multi-provider router with automatic fallback, circuit breakers, per-request cost budgeting, prompt caching, and token/rate governance. No single-vendor lock-in.
Quality
A scored evaluation harness that catches retrieval-quality regressions before release. RAG accuracy treated as a measurable, testable property.
Production reality
Subscription billing, org-scoped multi-tenancy, object storage, observability, disaster-recovery runbooks, pre-launch security hardening. Deployed and operated on real infrastructure.

The platform — five more, one codebase

Built solo on a shared multi-tenant core

Five additional production-grade AI SaaS products run on one shared Elixir platform, each a thin shell over hardened infrastructure. They're pre-launch, so the products stay dark — but the engineering doesn't. These are the primitives, described without the market. Full walkthroughs under NDA.

SYS · CLASSIFYUnder NDA

Deterministic classification & scoring engine

Rule-based, with a full auditable decision trail — chosen over an LLM where traceability and defensibility matter more than flexibility.

SYS · MEDIAUnder NDA

Multi-stage generative-media pipeline

Brief → storyboard → asset generation → voiceover → render, orchestrated through resilient background workers across several media providers.

SYS · GRAPHUnder NDA

Entity & relationship knowledge graph

A graph store layered over Postgres for entity-and-relationship intelligence, kept in sync from the primary system of record.

SYS · INGESTUnder NDA

Resilient ingestion & extraction

Browser-automation ingestion for dynamic, JavaScript-heavy sources, plus multi-format document extraction/OCR and normalization.

SYS · COREShared platform

Shared multi-tenant core

Unified auth, billing, RAG, chat, search, and content — so each new product ships as a thin shell, not a rebuild. Improvements compound across every app.

DEEPER PROOF

Architecture walkthroughs under NDA

Under mutual NDA I walk my own proprietary systems — real screenshots, workflows, architecture. I never use client work as public proof.

Where I can help

Engagements I take on

Fixed-scope pilots or weekly retainers, not open-ended hourly. I lower the risk of a narrow, high-value outcome.

Production RAG & document intelligence

Reliable retrieval, source citations, evaluation harnesses, anti-hallucination controls — audits through full builds.

AI feature integration for existing SaaS

Add an AI capability without breaking your auth, data, search, or UX. I build the whole system around the model.

Elixir / Phoenix build & rescue

Backend, LiveView, Ash, Postgres, OTP, deployment — for teams on the stack or moving to it.

Fractional technical partner / CTO

Architecture, delivery, and product judgment for founders who need someone who has been the PM, engineer, and operator.

Fixed-scope starters — low commitment, fast proof

RAG / retrieval-quality audit

A scored review of your retrieval, chunking, citations, cost, and failure modes — with a concrete fix roadmap. Fixed fee, days not weeks.

AI-generated code review & hardening

If your MVP was built fast with AI assistance, I review it for correctness, security, maintainability, and production readiness — and harden what needs it.

LLM API / AI-feature integration sprint

One reliable AI feature into your existing product — retrieval, structured outputs, or tool-calling — wired into your data, auth, and UX.

AI SaaS launch-readiness audit

A pre-launch pass over your product — reliability, security, cost, and data — closing the gap between “code-complete” and “ready for real users.”

How I work

Judgment first, tools second

My edge isn't typing speed. It's taking an ambiguous product problem, designing the system, building the full stack, integrating AI responsibly, deploying it, measuring cost and usage, and iterating — the whole arc, owned by one person who has also carried a P&L.

I use AI-assisted development as leverage, with human ownership of architecture, review, testing, and production quality. The AI writes code; I own whether it's correct, fast, secure, and maintainable under load. That distinction is the entire job.

Several of my own products are pre-launch, so everything here is deliberately abstracted — that's the same discretion I'll apply to your work. Deeper technical walkthroughs happen under mutual NDA, never in public.

Before the code — 18 years operating

Operator & founder track record

2024 —
Co-Founder & CTO, Vidhi · production legal-tech AI platform, built and shipped solo
2022 – 23
Head of Gaming Operations & Analytics, Bzinga · ran ops for a real-money gaming platform
2018 – 20
Planning & Analytics Leader, Reliance Brands · ~INR 800mn cost/inventory optimization; self-serve analytics automation
2012 – 16
Co-Founder & COO, Asankhya Retail · built an e-commerce venture; tech, supply chain, P&L, ~70% repeat customers
2006 – 12
Category & merchandising leadership · Flipkart/Letsbuy, Next Retail, HyperCITY — 200%+ category growth

MITx MicroMasters, Statistics & Data Science (MIT) · Independent Directors Certification, IIM Bangalore · MBA · B.Tech, Computer Science

Let's talk

Have a system that needs building — or rescuing?

Best fit for production RAG, AI features in real products, and Elixir/Phoenix work. Tell me the outcome you need; I'll tell you the narrowest first step to prove I can deliver it.

Email me →
Email   [email protected]
Based   Bangalore, India · remote worldwide