AI-native software. Built from Kolkata. Delivered everywhere.
SourceForge Software Services Pvt Ltd is a 12-year-old software company that designs, builds and supports AI-native business systems — ERP, POS, hospitality, education and marketing. We work with mid-market businesses across 11 countries, and we run on the same products we sell.
- 12+ years in software
- 350+ projects delivered
- 11 countries served
- Anthropic · OpenAI · Microsoft · AWS
Official partner across the AI and enterprise stack
From a custom-dev shop to an AI-native product company.
SourceForge began in 2013 as a software services shop in Kolkata. Twelve years and three hundred projects later, we're a product company with nine flagship platforms, four service lines, and a single operating principle — AI does the production work, humans do the judgment work. Here's the arc, in seven moments.
Founded in Kolkata
Started as a software services shop building custom Web and Windows applications for SMBs across India.
First product — ERP
Released the first version of SourceForge ERP. Five customers in year one. Built the on-prem deployment muscle that we still use today.
Microsoft Business Central partner
Became a Microsoft partner for Dynamics 365 Business Central. Custom extensions, implementation, and integration work for mid-market manufacturers and retailers.
First vertical product
Launched Gold & Jewellery ERP with live gold-rate-linked POS, hallmarking and AML. The first proof that vertical-native software beats horizontal ERP for the right buyer.
AI-native pivot
Brought Anthropic Claude into the heart of every product roadmap. AI Assistant for Business Central, OCR Payable Agent and Language Lab shipped within twelve months.
Marketing OS + CognitiveIQ
Released Marketing OS (60+ Claude-powered agents) and CognitiveIQ (three-test assessment platform). Crossed 11 countries of customer deployments.
Where we stand today
Nine flagship products, four service lines, 350+ projects delivered, 11 countries. The next chapter is global scale of the AI-native portfolio.
Six engineering principles that show up in every product.
Beliefs are cheap; principles cost money. These six are expensive to hold and worth every rupee. They're visible in our code, our SLAs and the way our products speak to their users.
Claude for text. OpenAI for media.
We chose one model family for every text task — analysis, generation, classification, narrative — and another for media. The discipline cuts brand-voice drift, doubles prompt-cache economics, and lets us reason about cost the way we reason about latency. Every line of agent code in our products honours this rule.
On-prem-ready, by default.
Every product runs in Docker. Every product can be deployed to AWS, to a VPS, to a school's server room, or to a customer's air-gapped DC. Host-specific config lives separately from app code; application code is portable. The customer chooses the host — never us.
Real SLAs, not theatre.
Every product page on this site quotes a real uptime SLA, a real P1 acknowledgment window, a real performance number. We publish them because we measure them. If we can't measure it, we don't claim it. "Best-in-class" and "world-class" are not numbers.
Honest copy, editorial tone.
We don't say "amazing" or "impressive." We don't use exclamation marks. We don't capitalise every noun. The reader is treated as an adult who can read prose. The same rule applies to AI narratives our products generate for end users.
Multi-tenancy from line one.
Every product that touches more than one customer uses Postgres Row-Level Security and a workspace primitive. "We'll add tenancy later" is a lie we never tell ourselves. The database schema, the RBAC, the audit log all start multi-tenant on day one.
Humans supervise. AI produces.
Nothing customer-facing goes out without human approval — unless the customer explicitly sets the auto-send rule. AI does the production work; humans do the judgment work. That ratio shifts cost without shifting accountability.
"We didn't pivot to AI. We pivoted to honesty."
When Claude crossed a threshold in 2023, we had a choice — bolt it on to the products we already sold, or rebuild the products around it. We chose to rebuild. It cost us a year of revenue and most of a roadmap, but it earned us the right to say AI is in the architecture, not in the marketing.
Twelve years in, what I'm proudest of isn't the products. It's the way they're built — multi-tenant from line one, on-prem-ready by default, real SLAs we publish because we measure them, and copy that reads like an adult wrote it for another adult. That's the company. That's what we hire for.
— Indrajit Banerjee, Founder & Managing Director
Nine flagship products. Four service lines.
Each product is a real shipping platform with paying customers — not a pitch deck. Each service line is a real practice with billable engineers — not a referral funnel.
SourceForge ERP
Cloud ERP — finance, inventory, sales, manufacturing.
Open pageGold & Jewellery ERP
28 jewellery-native modules — live gold-rate, hallmarking, AML.
Open pageHotel Management ERP
Front-desk to F&B, channel manager, owner dashboard.
Open pageRestaurant POS
Five offline-first apps — POS, KDS, Captain, Dashboard, Admin.
Open pageOCR Payable Agent for BC
Email-to-posted vendor invoice with three-way matching.
Open pageAI Assistant for Business Central
Claude-powered conversational interface inside BC.
Open pageLanguage Lab
AI English lab with Whisper ASR + phoneme scoring.
Open pageMarketing OS
60+ Claude agents running a marketing department.
Open pageCognitiveIQ
Three-test assessment platform with QR-verified certs.
Open pageCustom engineering when a product doesn't fit.
Measured. Published. Stood behind.
Six phases. Same playbook. Every product, every customer.
The same delivery sequence we run for a 50-user SourceForge ERP rollout, a 200-outlet Restaurant POS deployment, or a 30-college Language Lab programme. Tuned to scope; never reinvented.
Discovery
We listen first. Current systems, pain points, future state, regulatory boundaries — written into an SOW before anyone writes code.
Design
Architecture first, schema second, UI third. We refuse to start UI before the data model is signed off in plain English by the customer.
Build
Two-week sprints, demo at the end of every one, customer-side QA in parallel. No "surprise reveals" three months in.
Migrate
Data extraction from the legacy system, mapping, parallel-run validation, cutover plan. The riskiest hour of any project — we rehearse it.
Go-live
Phased cutover during a low-traffic window. Consultant on-site for the first 72 hours of every go-live, every product, every customer.
Hypercare
Four weeks of daily standups, then SLA-based managed support and a quarterly business review. We stay long enough to see the system breathe.
Six practices. One operating principle.
Founder-led; lean by design. Every practice has a lead with at least seven years in the discipline. AI doesn't replace the team — it lets the same team ship at the cadence a much larger team used to.
Want to join?
We hire on principle alignment first, skill second. Open roles across engineering, AI and customer success.
The technology stack we're officially accredited on.
We're an API-tier partner with Anthropic and OpenAI, a Microsoft Business Central partner for AL extensions, and an AWS cloud partner. Accreditations matter — they mean we get model access, support and roadmap visibility our customers benefit from.
We don't collect partnership logos for the website. Each badge represents a real working relationship with shared customer commitments.
Where we are
Headquartered in Kolkata. Delivering across 11 countries — India, the GCC, Southeast Asia, the UK and North America. On-prem and cloud deployments, customer-side go-lives wherever they sit.
- 1/3 Bangur Avenue, Block D, 4th Floor, Kolkata 700055
- +91-98306-33-555
- info@sourceforge.in
Press & news
Product launches, partner announcements, customer go-lives and our writing on AI-native business systems. Everything in one place, dated and sourced.
Press enquiries: info@sourceforge.in
30-minute discovery call — no slide deck.
Tell us what you're trying to do — a product to roll out, a service engagement to scope, or a partnership to explore. We'll read it before the call, come with two or three options, and stay on the time we booked.
- 30-minute discovery call
- Two-three options on the table
- Written follow-up the same day
- Real engineer on the line — not a sales rep