Forge & Function — AI Engineering Studio

We design & ship production-grade AI systems.

Intelligent systems, full-stack platforms, and automation for teams worldwide — from first architecture to production and beyond.

Forge & Function is a founder-led AI engineering studio. We design and build production systems — the kind that run every day, under real load, with real money and real operations depending on them.

What we build

AI Systems & Agents — Multi-agent platforms, RAG pipelines, and LLM applications that do real work.
Full-Stack Platforms — Production web apps and SaaS, designed and built end to end.
Automation & Trading Systems — Algorithmic trading, market data, and workflow automation.
Data & Infrastructure — Pipelines, databases, and cloud reliability that hold up under load.

How we work

Discovery — We scope the problem, constraints, and success metrics — and agree on what 'done' means before any code is written.
Architecture — We design the system, choose the stack, and break the build into clear, demoable milestones.
Build — We ship iteratively with regular demos, so you see real progress and steer the direction as we go.
Ship & Support — We deploy to production, hand off with full documentation, and support what we built.

Who runs the studio

Forge & Function is founded and led by Shubham Mandal, Founder & Principal Engineer. You work directly with the engineer who designs and builds your system.

Shubham is an AI engineer and infrastructure architect who came to AI the long way around — through the unglamorous discipline of keeping enterprise systems running.

As a Senior Infrastructure Analyst at DXC Technology, he has spent 3+ years administering mission-critical Oracle environments on Unix/Linux: RMAN and Veritas backups at ~99.9% success, performance tuning through AWR and ASH, Data Guard failover drills, and automation that cut manual DBA workload by roughly 60%.

By the numbers

13 — Systems built
7 — Domains served
3+ — Years in infrastructure
100% — Production-focused

Selected work

DIP Engine — Beauty-Product Intelligence Pipeline — A storage-free ingestion pipeline that turns a retailer's own product APIs into clean, continuously-refreshed beauty-product data for competitive-intelligence dashboards. Result: 0 per-product AI cost. Collapsed a ten-phase AI/ELT stack into a five-table, two-step engine that is cheaper and simpler to operate.
IOP AI — Intelligent Operational Platform — Multi-agent supply-chain & manufacturing intelligence platform with departmental RBAC. Eliminated stockouts via multi-plant procurement visibility & SLOB reduction.
Layers — Private Loan Manager — Private two-owner app to track borrowers, loans, payments, collateral, and account balances with reminders. Replaced scattered manual notes with a single source of truth for balances and overdue status.
HoneyMoon AI — Wedding & Event Planning Assistant — A bilingual AI consultant that plans UAE weddings by matching couples to real vendors across 20 categories conversationally. Result: 20 category agents, zero hallucinated vendors. Eliminated vendor hallucination by separating LLM narrative from database-served facts.
Saloo.live — Barbershop Booking & Queue Platform — India-first barbershop platform for appointments, live walk-in queues, and online advance payments. Replaces phone-and-walk-in chaos with real-time queue visibility and estimated wait times.
AI Chatbot+ — White-label RAG chatbot that crawls any website and turns it into an intelligent sales copilot. Deployed live on Vercel with full chatbot functionality in production.
StreamRoom — Private, invite-only watch party platform with real-time video sync and live chat. Enables private, synchronized watch parties across any distance.
Cloview — AI Trading Dashboard — Multi-agent AI trading analysis platform with real-time MT5 market data and smart money detection. Consolidates 14 independent analyses into a single actionable signal with confidence score.
MentoroidAI — Adaptive Learning Engine — Rule-based adaptive learning engine that personalizes K-10 learning paths using real-time mastery, stress, and confidence signals. Architecture designed for sub-200ms end-to-end decision latency, with the adaptive node decision constrained to under 50ms.
Tea Packaging Optimization Platform — Computes the cheapest physically loadable pouch-to-container packing plan for tea export shipments. Result: 69% container utilization, up from 36.9%. Container utilization improved from 36.9% (sequential pipeline) to ~69% on the reference shipment.
NewsPulse — Beauty-Product News Intelligence Engine — A backend engine that turns scattered beauty-industry news into cited, catalog-linked product insights for a cosmetics market-intelligence platform. Result: 79% brand match, up from 28%. Surfaced Korean beauty coverage that direct scraping missed entirely (0 vs 13 of 15 articles).

What we believe

Ship it or it doesn't count — A system that isn't in production isn't finished. Every engagement ends with deployed, documented, handed-over software — not a prototype and a slide deck.
Grounded, not guessed — AI output is only as trustworthy as the data behind it. We serve facts from real sources and keep the model's job narrow, so answers hold up to scrutiny.
Design for the bad day — Backups, failover, rate limits, edge cases. We plan for the failure modes up front, because that's what separates a system from a script.
Direct, not layered — You talk to the engineer designing your system. No account managers, no telephone game, no context lost between the brief and the build.

Common questions

Do you build custom machine learning models, or use existing ones?

We build applied AI systems on top of existing foundation models — multi-agent orchestration, retrieval pipelines, and grounding models in real data so their output can be trusted. We do not train bespoke models from scratch, and we do not offer fine-tuning as a standalone service. Where a problem is better solved deterministically we say so rather than reaching for a model: the Tea Packaging optimiser scores ~15,000 configurations with exact maths, and DIP Engine removed an LLM cleaning layer entirely once we measured it as redundant.

How do you handle data security and privacy?

Security is designed into the architecture rather than added afterwards. In practice that means row-level security in the database so each tenant and role can only read its own rows (Layers, Saloo.live); departmental RBAC isolating what each agent can access (IOP AI); grounding models in your data so they cannot invent facts about your business (HoneyMoon AI, NewsPulse); and credentials kept in environment configuration, never in source. Systems are built in your cloud accounts where you have them, and infrastructure and data are handed to you on delivery. We hold no compliance certifications such as SOC 2 or ISO 27001 — if your procurement process requires them, say so early and we will tell you honestly whether we can meet it.

Who owns the code and the intellectual property?

You do. Source code, infrastructure and documentation transfer to you outright on delivery, and we retain no licence over what we build for you.