We train your people, set up your AI centre of excellence, and build systems that run inside your own boundary and stay after we leave.
Nothing here awards a contract or sanctions a payment. Each step ends with an officer deciding, and a record that explains the decision.
In brief
EigenSpark works with central and state public sector enterprises, boards and authorities. We start by training your officers and engineers, then stand up a centre of excellence they run. We build alongside them: tender evaluation, contract review, grievance handling, plant and asset work, and the back office. Everything runs inside your own boundary, and every step leaves a record that answers a vigilance query or an audit para without a scramble.
What we cover
Approvals, vigilance, audit and data protection are designed in from the first week.
The starting position
Capability that left with the last vendor, a long approval chain, and every unit doing it differently.
A pilot is delivered, it works, and nobody inside can extend it. The second use case needs a fresh procurement.
An officer who has to sign cannot use a recommendation that hides its reasoning. Every output needs its source, its rule and its owner.
Legacy applications, a half-finished ERP, spreadsheets held by one person, and paper. Most public sector use cases start as a data engineering job.
What your senior people know about the plant, the file and the regulation is in no manual. Capturing it has a hard deadline.
How an engagement runs
Three tracks at once: strategy, training your people, and standing up the centre of excellence.
A readiness read across the functions you want changed, at headquarters and at representative units. Then the strategy: a ranked opportunity map, a sequenced roadmap, and the governance that says who approves what, funded against the MoU parameters.
Board and executive fluency so a sponsor can judge a proposal and challenge a vendor. Business tracks for the functional teams. Technical depth for IT and the analytics group. Not all of it is AI: SQL, data engineering, cloud and cybersecurity run alongside.
Use cases identified with the business, built by your own teams with our engineers alongside them, reviewed monthly. The measure of this track is whether the enterprise ships the next use case without us in the room.
Use cases
Procurement and contracts, then vigilance and audit, then citizens and staff, then the back office.
Bids read against the criteria you published, returned as a matrix with every deviation linked to the clause it fails. It evidences; the committee awards.
Contract Risk Analyser →Contracts read into a structured record: obligations, milestones, penalties, renewals and indemnities, with every change from your standard tagged by severity.
Contract Risk Analyser →Spend across GeM, CPPP and legacy records in one view, with duplicate vendors, rate variance between units and repeated single-bid categories surfaced.
AI & Data Strategy →Every observation held with its origin, the unit responsible and its reply history, and draft responses assembled from the source documents.
Continuous Compliance Monitor →Transaction, approval and vendor data read for patterns worth a look: splitting below thresholds, unusual approval sequences, vendors sharing details.
Agentic AI →The request parsed, the records located across the file system, and a draft reply assembled with every fact carrying the document and page it came from.
Generative AI →Complaints classified, routed and tracked to closure, with recurring causes grouped so the process gets fixed. Ageing is visible to the officer who owns it.
Document Intelligence Engine →An assistant that answers routine questions from your published circulars and approved material, cites the source, and hands anything else to a person.
Generative AI →Decades of office orders and policy notes made answerable in plain language, with the current version identified and superseded ones marked.
Generative AI →Figures pulled on a schedule, variance against target explained in plain language, and the board pack assembled from one set of numbers.
Data Engineering & Platforms →Bulk hiring screened against the advertised criteria with the reasoning recorded, and succession modelled for the roles where retirements are already visible.
AI & Data Strategy →The judgement of retiring officers moved into a searchable, cited record. Interviews, manuals and case history become an assistant that shows its source.
Generative AI →The engineering half
Four layers that decide whether a system gets approved, runs inside your boundary and survives an audit.
Most PSU use cases are a data engineering answer before they are a model answer, and the roadmap says so.
The record stays where it is. We connect to it.
Every recommendation carries its input, the rule it applied and the person it goes to, because an output that cannot be explained adds risk to a file.
Open-weight models on your own hardware or your sovereign cloud, so a change of vendor or of policy does not strand the system.
We also run the training that goes with it: SQL and data engineering, cloud, enterprise systems and cybersecurity, alongside the AI tracks. A team that cannot query its own data cannot specify a model either.
What we offer public sector teams
Most enterprises start with training, and end up using three of the four.
Where most PSU engagements start.
Strategy, governance and the centre of excellence.
Where a single workflow starts.
The teams that own this work.
Working in a specific function? See how we help Procurement teams.
Security and controls
The officer decides. It runs inside your boundary. The trail is complete and nothing is hidden.
Every system here recommends, evidences and records. Award, sanction, disciplinary and disclosure decisions stay with the authority that holds them, and the system is built so that boundary is visible in the file.
On your own hardware or your sovereign cloud, with open-weight models where the data cannot leave. Where a hosted model is used at all, what is sent to it is agreed and recorded before a line is written.
Every action carries a user, a timestamp and a before-and-after, and every output carries the source it was derived from. A vigilance query, an audit para or an RTI is answered from the record by your own team.
Source, infrastructure definitions, documentation and model weights where the model is ours to hand over. The intent of every engagement is that the enterprise can extend the system without a fresh procurement.
FAQs
Yes, and for most enterprises we would recommend it. Open-weight models hosted on your own GPUs or your sovereign cloud handle the large majority of the use cases on this page at a quality that holds up in production. Where a hosted model genuinely adds something, what is sent to it is agreed and documented before build, and your CISO makes that call, with it written down.
By carrying its reasoning. Every output records the input it was derived from, the rule or criterion applied, the confidence where that applies, and the officer it was placed before. Nothing is awarded, sanctioned or disclosed by a system. When a query arrives, your own team pulls the trail; nothing has to be reconstructed and we do not need to be involved.
The engagement is structured so your teams build. Our engineers sit alongside them, the code and the documentation are yours throughout, and the review each month asks one question: can this team ship the next use case on its own. Where the answer is still no, that is the thing we work on next.
Through your normal route. We supply on GeM where the category exists and through the tender route where it does not, and the engagement is scoped so it can be evaluated against published criteria. We have no objection to Integrity Pact conditions, and we will say early where a requirement is written in a way that only one vendor can meet.
That is the work. The first phase of most engagements here is data engineering: getting to one reliable record out of the applications, the partially rolled out ERP, the departmental spreadsheets and the scanned files. We would rather tell you that in the assessment than discover it in month four.
No. Central and state enterprises, boards, corporations and authorities. The governance language differs, the shape does not: an approval chain that has to hold, data spread across systems, a capability gap, and an audit that will eventually ask.
No, and the parts that are not are usually what makes the AI work. Data engineering, integration, access control and workflow are ordinary engineering and they are most of the effort. The same is true of the training: alongside the AI tracks we run SQL, data engineering, cloud, enterprise systems and cybersecurity.
An assessment across the functions you name, at headquarters and two or three representative units, producing a readiness score, a ranked opportunity map and a sequenced roadmap with go and no-go points. In parallel the leadership programme runs, because the decisions in phase two are better when the people making them have seen the technology work.
Related
The two pillars this runs through, the products behind it, and the teams that own it.
The tiered curriculum, and how a cascade is run.
→ PillarStrategy, governance and the centre of excellence.
→ ProductTender and contract scrutiny against your own criteria.
→ ProductAudit paras, checklists and the evidence behind them.
→ FunctionThe function that owns tenders, vendors and spend.
→ IndustryThe plant and mine side, for CPSEs in the sector.
→Name one function and one unit. We call you within 48 hours and go through the use cases worth doing there, the data work each one needs first, and what a first phase would take.
Talk to us