Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Yes, you can build a useful private assistant for engineering records without sending those records to a hosted language model. The design described here collects Git history and Monday.com project data into SQLite, asks a local model such as Ollama running qwen2.5-coder:7b to generate SQL, executes that SQL locally, and has the same local-model workflow explain the returned rows. It works especially well when your questions involve explicit fields—authors, dates, statuses, assignees, branches, and deadlines—rather than paragraphs of unstructured text. The result is a practical way to ask questions such as “Who committed the most to our main repo this month?” or “What project tasks are still in progress for the web team?” without manually searching terminal output and project boards.
What exactly did I build?
The assistant has two collectors and one query path. A Git collector runs git log across repositories and writes commits, file changes, branches, and tags to a SQLite database. A project-management collector calls Monday.com’s GraphQL API and stores boards, items, and subitems. The conversational layer does not answer from a preloaded chat memory: it turns a natural-language question into SQL, reads the current database rows, and summarizes those rows.
As an Amazon Associate I earn from qualifying purchases.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →That distinction is important. The language model is being used as a translator and explainer, while SQLite remains the system of record for the structured information. The implementation is described by Pouria Zandakbari in the original project article. It is a first-person account of one design, not an independently audited privacy or accuracy study.
The end-to-end flow
- Collectors periodically read Git repositories and the project-management API.
- They normalize the results into related SQLite tables.
- A user asks a question in ordinary language.
- The system searches for relevant names and values, then gives those candidates to the local model.
- The model writes SQL that targets the SQLite schema.
- SQLite executes the query and returns rows or aggregates.
- The local model explains the result in plain language.
For example, “What files were changed for the last campaign launch?” needs a relationship between a campaign task, its branch, and the files recorded in Git. The answer is not produced by semantic similarity alone; it depends on the data model and the joins the generated SQL selects.
#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
Why use SQLite and Text-to-SQL instead of a vector index?
I chose relational querying because commits and tasks already have fields that databases understand. Filtering by a date range, grouping by author, counting file changes, or selecting items whose status is “In Progress” is naturally expressed as SQL. This is a design rationale for this workload, not proof that Text-to-SQL is always more accurate than retrieval-augmented generation.
| Question shape | Why relational querying fits | What must be modeled |
|---|---|---|
| Who committed the most this month? | Date filtering, grouping, and counting are explicit operations. | Commit author and timestamp, plus the repository scope. |
| What files changed for a launch? | A join can connect a task or branch to per-file additions and deletions. | Task names, branch names, commit records, and file-change rows. |
| Which web-team tasks remain in progress? | Status, department, and assignee are structured filters. | Board, item, subitem, status, department, and assignee fields. |
Document search has a different shape. A separate Local-RAG-Assistant project demonstrates a local document workflow built around Ollama, Qdrant, embeddings, and a document store. That architecture is better suited to finding relevant passages in notes, specifications, or support documents. It does not establish a general winner against the SQLite approach; the data shape and questions should determine the choice.
Recommended Free Tools
How are Git records collected?
The Git collector walks the repositories I choose and uses their history as the source for development activity. Each commit record includes the author, date, message, repository context, and associated branch or tag information. File-level records retain the files touched by a commit and the insertion and deletion counts for each file.
Why file-level history matters
A commit-only table can tell you who committed and when, but it cannot answer “What files changed?” reliably. Storing one row per changed file allows the assistant to aggregate changes for a commit range, a branch, or a task-linked campaign. It also lets a response distinguish a large number of edits in one file from a broad change spread across many files.
The collector is a snapshot-and-refresh component, not a magical view into every repository. Repositories must be available to the process, and the refresh schedule determines how current the SQLite copy is. If a commit has not been collected yet, the assistant cannot report it.
How does the Monday.com collector fit into the database?
The second collector queries Monday.com’s GraphQL API and stores boards, items, and subitems. The fields described in the build include status, assignee, department, and deadline, along with the names and identifiers needed to connect records to other tables.
Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
GraphQL gives the collector a way to request the board fields needed for analysis instead of treating the project system as a screen to scrape. API credentials, permissions, pagination, and refresh errors remain operational concerns for the collector. The described implementation uses Monday.com; Jira and Linear are suggested as possible adaptations, not systems demonstrated by the project.
A compact view of the stored entities
| Source | Stored records | Useful fields | Typical questions |
|---|---|---|---|
| Git | Commits, file changes, branches, tags | Author, date, message, file path, insertions, deletions | Who changed what, when, and in which repository? |
| Monday.com | Boards, items, subitems | Status, assignee, department, deadline, names and identifiers | Which tasks are pending, assigned, or overdue? |
| Relationship layer | Branch-to-task mappings | Branch name, task name, board name | Which code changes belong to a project task? |
How do branches become linked to project tasks?
The explicit branch_task_map is one of the most important design decisions. It records which branch corresponds to which task and board name. Without this table, a language model can guess that two names look related, but it has no dependable relational key for joining project state to code history.
In practice, the mapping can be populated when a branch is created, when a task identifier is included in a branch name, or through a reviewed import process. The exact automation is an implementation choice; the essential point is that the relationship is stored as data. Once it exists, a query about a named campaign can find the matching task, follow the mapped branch, select its commits, and expand those commits into file changes.
What happens when I ask a question?
The assistant uses several stages rather than handing the entire database to the model as conversational context.
- Receive the question. The user asks in natural language, such as “What tasks are pending for the next launch?”
- Discover likely values. The application searches across tables for matching identifiers and values—project names, branch names, people, statuses, or repository names.
- Generate SQL. The local model receives the schema context and the discovered candidates, then proposes a query.
- Execute against SQLite. The application runs the SQL and obtains rows, counts, or grouped results.
- Retry empty results. If the first strategy returns zero rows, the implementation tries alternative keyword strategies. This improves discoverability but cannot resolve every naming mismatch.
- Summarize the returned data. The model explains what the rows show instead of inventing an answer from memory.
The author summarizes this design plainly: “The LLM doesn’t store or memorize my data. It just translates my question into SQL, runs it, and explains the results.” That is a description of the intended workflow, not an independent audit of every component that might handle the data.
Which questions does this design answer well?
The strongest questions combine filters, joins, and aggregates over fields already captured by the collectors.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
- “Who committed the most to our main repo this month?” Filter commits by repository and date, group by author, and order by the count.
- “What files were changed for the last campaign launch?” Resolve the campaign task, follow its branch mapping, select relevant commits, and list their file rows.
- “What project tasks are still in progress for the web team?” Filter the project records by department and status, then return task names, assignees, and deadlines.
- “What tasks are pending for the next launch?” Use the launch identifier and status fields to isolate work that has not reached completion.
These are not merely keyword lookups. The useful answer depends on choosing the right date field, repository, board, status value, and relationship path. A generated query that omits one of those constraints can be syntactically valid and still answer the wrong question.
How can I make generated SQL safer?
The reported workflow demonstrates generation, execution, and retry logic. For a production version, I would put explicit controls between the model and the database.
- Use a read-only SQLite connection for assistant queries.
- Allow only
SELECTstatements and reject data-definition or write operations. - Allow-list tables and functions exposed to the model.
- Apply row and execution-time limits so an accidental broad join cannot overwhelm the process.
- Log the question, generated SQL, execution error, and returned row count for review.
- Show the SQL or a concise query explanation when an answer affects planning or reporting.
- Return “no matching rows” distinctly from a failed query; those cases require different fixes.
These safeguards do not make generated SQL infallible. They reduce the blast radius of a bad query and make it possible to diagnose whether the problem is a missing collector record, a wrong keyword, a schema misunderstanding, or a legitimate empty result.
How local is a “private” assistant?
The described setup runs Ollama with qwen2.5-coder:7b on Apple Silicon. The intended inference path and database processing are local, and the source does not name a specific Mac model, memory size, benchmark, or minimum configuration. It therefore supports the category “Apple Silicon MacBook” as a setup context, not a hardware recommendation.
Local inference should not be read as a network-audit result. The project-management collector must contact Monday.com’s API when it refreshes data, and operating-system services, backups, integrations, or other tools may have their own network behavior. The available account describes the intended data flow; it does not prove that every surrounding service sends nothing externally.
There is also no published speed or accuracy benchmark for this particular assistant. The visible sample counts in the original article are generalized and rounded, so they should not be used to estimate throughput, database size, or model quality.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
What is the practical difference from a local vector-RAG assistant?
A vector design converts document chunks into embeddings and retrieves semantically similar passages. The SQLite design preserves typed values and relationships, then asks the model to express the question as SQL. Both can be local, but they solve different retrieval problems.
| Decision axis | SQLite plus Text-to-SQL | Vector RAG |
|---|---|---|
| Best source shape | Commits, tasks, statuses, dates, assignees, and other structured records | Notes, specifications, messages, and other document text |
| Typical query | Filters, joins, counts, rankings, and date ranges | Find passages that discuss a concept or answer a question semantically |
| Relationships | Must be represented explicitly, such as branch_task_map |
May be inferred from nearby text, metadata, or retrieved chunks |
| Freshness work | Refresh collectors and keep schema values synchronized | Re-embed changed documents and maintain the vector store |
| Main failure mode | Incorrect SQL, missing joins, stale rows, or unmatched identifiers | Irrelevant chunks, missed passages, or weak grounding |
| Processing location | Can keep the database and model on the local machine, subject to surrounding services | Can also be local, as shown by the Local-RAG-Assistant example |
If your next question is “Which paragraph in the incident review explains the rollback?”, document retrieval is appropriate. If it is “How many commits touched the authentication service after the task entered review?”, typed records and SQL are usually the clearer foundation.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How would I build a similar assistant?
A repeatable implementation starts with the data contract, not the chat interface.
- Define the questions first. Write representative questions and list the fields each answer needs. This exposes missing dates, identifiers, or relationships before any model is involved.
- Build the Git collector. Run
git logover the selected repositories and persist commits, branches, tags, and per-file insertion/deletion records. - Build the project collector. Query the Monday.com GraphQL API for boards, items, subitems, and the status, assignee, department, deadline, and naming fields required by your questions.
- Design stable keys. Preserve source identifiers and names, and create the
branch_task_maprelationship rather than relying on fuzzy matching at answer time. - Add refresh and reconciliation. Decide how often each source is collected, how updates are detected, and how deleted or renamed records are handled.
- Add value discovery. Search for likely names and identifiers before prompting the model, and provide those candidates with the schema.
- Constrain SQL execution. Use read-only connections, allow-lists, limits, and logs before exposing the assistant to real project data.
- Separate answering from summarizing. First obtain and inspect rows; only then ask the local model to explain them.
Keeping those stages separate makes troubleshooting possible. A wrong answer can be traced to collection, mapping, discovery, SQL generation, execution, or summarization instead of being dismissed as a vague “AI error.”
Free tools Windows power users keep installed
One-click scans. No signup required.
What should I test before trusting the answers?
Create a small evaluation set from questions your team actually asks, and verify the expected rows manually. Test boundary dates, renamed branches, duplicate task names, empty result sets, missing assignees, and tasks with multiple subitems. Check that a repository filter does not silently include another repository and that a status value is interpreted exactly as it appears in the project system.
Do not treat an unrelated benchmark as evidence about this build. Alan McDonagh’s capstone report, published May 31, 2026, describes a 20-question evaluation for a separate project; it is useful context, not a performance result for this assistant. The original build likewise provides no independently published statistic or reproducible speed measurement.
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
When is this architecture the right choice?
Choose the SQLite-and-Text-to-SQL pattern when your source systems expose durable fields and your questions involve joins, filtering, aggregation, or time windows. Invest early in identifiers and relationship tables, because the quality of those links matters more than conversational phrasing.
Choose a vector-RAG pipeline when the central value is locating meaning inside unstructured documents. Many teams will eventually need both: relational tables for authoritative activity and task state, plus a document index for explanations and narrative context. The key is to route each question to the representation that actually contains the evidence.
This build is compelling because it treats the language model as an interface to well-modeled engineering data rather than as a replacement for the database. Its privacy advantage is an intended local inference path, while its practical limits are data freshness, relationship quality, generated-query reliability, and the behavior of every surrounding service.
Frequently Asked Questions
Does local inference guarantee that no project data ever leaves the computer?
No. The model and SQLite processing can remain local, but refreshing Monday.com data requires an API connection, and backups, integrations, or operating-system services may have separate network behavior. Verify the complete deployment and data-flow boundary rather than relying on the word “local.”
Can I connect Jira or Linear instead of Monday.com?
The same collector-and-schema pattern could be adapted to them, but the described implementation only demonstrates Monday.com’s GraphQL API. You would need to map each service’s identifiers, status values, users, and hierarchy into your own tables.
Do I need a vector database for Git and task questions?
Not for the structured questions this design targets. A vector store becomes useful when you also need semantic search over documents such as design notes, retrospectives, or incident reports.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesHow much RAM does qwen2.5-coder:7b require for this project?
The account identifies Apple Silicon support but gives no minimum memory, model-loading profile, or benchmark. Treat hardware sizing as an environment-specific test rather than a guaranteed specification.
What should I do when the assistant returns no rows?
First determine whether the collector has refreshed the relevant source and whether the branch-to-task mapping exists. Then inspect the discovered keywords and generated SQL; retry logic can try alternate names, but it cannot invent a missing record or relationship.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
