Skip to content

Scope, fee and working days are published on every engagement

support@corephysixlabs.com 7702752958 Track an order
CorePhysix Labs Technology lab
Start a project
01 / WEB Web & App Engineering 5 engagements · from ₹24,999 02 / CLD Cloud & DevOps 5 engagements · from ₹12,999 03 / DAT Data & AI 5 engagements · from ₹14,999 04 / UXD Product Design & UX 5 engagements · from ₹14,999 05 / API APIs & Integrations 5 engagements · from ₹18,999 06 / SEC Security & Compliance 5 engagements · from ₹12,999 07 / QAT QA & Test Automation 5 engagements · from ₹12,999 08 / SUP Managed Support 5 engagements · from ₹8,999 09 / ADV Tech Advisory 12 engagements · from ₹100

Not sure where to start?

Book a short advisory session and let a specialist point you to the right engagement.

See advisory sessions Full catalogue

Specification DAT·104

DAT · Data & AI Defined scope Fixed fee

Retrieval Chatbot on Your Documents

An assistant that answers from your own manuals and policies with a citation on every reply, reducing repeat support questions.

Fee · GST incl.
₹44,999
Duration
22 working days
Deliverables
7

01 Overview

A general-purpose chatbot has never seen your documents, so it fills gaps about your business with guesses. We build a retrieval-augmented assistant over the material you supply: policies, product manuals, support articles or contracts. Content is split into chunks and embedded in a vector store, retrieval is tuned against a question set you sign off, and every reply cites the paragraph it relied on. If the answer is not in the documents, the assistant says so, and we test that behaviour before handover.

02 Deliverables

7 items

What you receive at handover

  1. D01 Pipeline that ingests PDF, Word, HTML and Confluence sources
  2. D02 Vector store whose chunk sizes and embeddings suit how your documents are written
  3. D03 Retrieval and answer service that cites a source in every response
  4. D04 Web chat widget or Slack bot, as you prefer
  5. D05 Evaluation set of 50 questions, with accuracy measured before and after tuning
  6. D06 Guardrails that handle off-topic questions and attempts at prompt injection
  7. D07 Cost model projecting monthly spend at your expected query volume

03 Outcomes

What should be true once it ships

  • Answers are grounded in your documents, each with a citation to verify
  • Fewer support tickets for the questions that come up most often
  • A measured accuracy figure rather than an impression from a demo

04 Process

How the 4 stages run

  1. Gather Documents are collected, deduplicated and cleaned
  2. Build We implement ingestion, retrieval and the answering service
  3. Evaluate Responses are scored on your questions, and retrieval is adjusted
  4. Roll out A pilot group first, then wider release once accuracy holds

05 Fit & inputs

A good fit for

  • Support teams fielding identical policy queries day after day
  • Companies with extensive internal documentation that is hard to search
  • Products whose onboarding relies on reading a long manual

What we need from you

  • The document set, with clarity on which versions are current
  • 50 real questions your users ask, along with the correct answers
  • Your choice of model provider, based on what your policies allow

These inputs are collected through the technical brief in your dashboard after payment. The duration is counted from when they arrive.

06 Answers

Questions about this engagement

Grounding answers in retrieved text and having the assistant decline when it lacks one cuts this back substantially, and we measure how often it still happens. Claiming it never happens would be false, so we do not.

No. We only use providers whose terms rule out training on customer data, or run a self-hosted model where your policy demands it.

Expect roughly ₹3,000 to ₹15,000 a month, driven mainly by how many queries you handle. We work out the figure for your volumes before you commit.

This is technology work delivered digitally — nothing is shipped. Work is delivered into systems you control within the duration stated above, and the delivery and refund policies set out what happens if a date is missed.

REL Related

Often considered alongside.

All Data & AI
DAT — Data & AI DAT·101

Postgres and dbt Analytics Warehouse

A Postgres or Snowflake warehouse with tested dbt models, so app, payment and ad data share one set of definitions everyone trusts.

Postgres or Snowflake warehouse provisioned with access roles

Fixed fee · GST incl.

₹39,999

Duration

25 working days

View scope
DAT — Data & AI DAT·103

Metabase Reporting Setup

Four Metabase dashboards on a governed model layer, built around leadership decisions so recurring questions no longer need an analyst.

Metabase set up with a connection to your warehouse or database

Fixed fee · GST incl.

₹14,999

Duration

9 working days

View scope
DAT — Data & AI DAT·105

Machine Learning Feasibility Study

A data audit and baseline model that show whether your data supports the model you want, before you commit budget to building it.

Audit of data volume, quality, labelling and class balance

Fixed fee · GST incl.

₹18,999

Duration

10 working days

View scope

GST included

₹44,999

Next step

Have a system that needs building, fixing or securing?

Explore the catalogue Describe the problem